Changes from 1.5beta0: ---------------------------------

- BUG FIX (apply1D.c): in setting colSize for 1D maps, when output
type and map type are different sizes
- API CHANGE (nrrd.h, map.c): nrrdHistoEq() now takes a new final
argument, "amount", which tell the extent to which the equalizing
correction should be applied.
- API CHANGE (simple.c, write.c): 3D grayscale images
(dimensions=(1,X,Y)) now "fit" in PGM, and are silently reshaped on
write
- API CHANGE (simple.c): nrrdPeripheralInit() and nrrdPeripheralCopy()
now return an int instead of void.
- API NEW (nrrd.h, defaultsNrrd.c): added nrrdStateGrayscaleImage3D,
which causes reshape of grayscale images to 3D (1,X,Y) on read
- API NEW (nrrd.h, methodsNrrd.c): nrrdKernelSpecSet() and
nrrdKernelParmSet() are ways of setting a kernel/parm pair and a
nrrdKernelSpec from one another

Changes from 1.4: ----------------------------------------

nrrd.h (and nrrd as a whole):
- API CHANGE: the evalN_f and evalN_d methods of NrrdKernel
now use size_t to indicate how many samples to evaluate,
instead of int.
- API CHANGE: there is no more nrrdBigInt.  It was stupid.
It should have been size_t all along, because the way I was
using nrrdBigInt is exactly the way that size_t is intended 
to be used.  Basically, all functions that linearly iterate
through all elements of a nrrd use a variable of type size_t.
On 32-bit architectures, using size_t means you use an
unsigned int instead of an unsigned long long int, which is
probably faster.  Associated with this is the demise of
NRRD_BIG_INT_PRINTF; its usage is replaced by AIR_SIZE_T_FMT
- API CHANGE: nrrdMinMaxClever() renamed to
nrrdMinMaxCleverSet()
- API CHANGE: nrrdApply1DLut() and nrrdApply1DRegMap() now
take two final int arguments:
  typeOut: what type should output be
  rescale: values from input nrrd should be affine-rescaled
           from nrrd range to lut/map domain.
nrrdApply1DIrregMap() uses the same two arguments.
- API CHANGE: nrrdHistoEq() can no longer allow nin==nout,
because of restrictions in nrrdApply1DRegMap().  This may
be fixed some day.
- API CHANGE: "longlong" new name for nrrdTypeLLong; 
"ulonglong" new name for nrrdTypeULLong
- API CHANGE: (but no one should notice): added dataFN[]
(data filename) to NrrdIO struct, basically so that
"unu make -h" can subvert _nrrdWriteNrrd().
- API CHANGE: (but no one should notice): there are no
more statically allocated strings in the NrrdIO, and 
no more NRRD_STRLEN_LINE.  Everything is dynamically 
allocated to allow arbitrarily long lines.  
- API NEW: added nrrdPad_nva, and nrrdSimplePad_nva
- API NEW: added nrrdUnquantize(), which provides a way of
approximating floating point data prior to a quantization.
- API NEW: the floating point type that nrrdSpatialResample()
uses for all intermediate value representation has been
generalized from float only to nrrdResample_t, which can
be set to float or double by controlling an #if in nrrd.h
- API NEW: added nrrdTypeMin[], nrrdTypeMax[], and
nrrdTypeNumberValues[], arrays which represent for the
*fixed*point* types the lowest and highest representable
values, and the number of different possible values.
- API NEW: added nrrdApply1DIrregMap(), for "colormapping"
with irregular univariate maps
- API NEW: added nrrd1DIrregMapValid(),
nrrd1DIrregAclValid() which may be useful for verifying
suitability of nrrds used in nrrdApply1DIrregMap().
- API NEW: added _nrrdFieldOnePerAxis[NRRD_FIELD_MAX+1],
which is non-zero for nrrdField_{sizes,spacings,etc.}:
those things for which you need as many values as there
are axes in the nrrd.
- API NEW: added nrrdTypeUnsigned[], which is non-zero for
things like unsigned chars and unsigned shorts
- FUNCTIONALITY CHANGE: tables can now hold 1D data, see
info for read.c and simple.c
- FUNCTIONALITY CHANGE: nrrdHistoEq() now permits learning
not the histogram ("nhistP"), but the transform used to
shift values ("nmapP") in a form suitable for
nrrdApply1DRegMap(); in fact that is what nrrdHistoEq() calls
itself.  This way one mapping can be applied to other nrrds.
- FILE FORMAT CHANGE: this was supposed to have been done
for teem-1.4: default nrrd magic is "NRRD0001", instead of
"NRRD00.01"
- fixed comment typos and spacing

nrrdMacros.h:
- API NEW: created NRRD_AXIS_CELL_POS(),
NRRD_AXIS_NODE_POS(), NRRD_AXIS_CELL_IDX(),
NRRD_AXIS_NODE_IDX(), which are special case versions of
NRRD_AXIS_POS() and NRRD_AXIS_IDX()

methods.c:
- BUG FIX: nrrdMaybeAlloc_nva() was not correctly calculating
the allocated size of the desired output- using 
nrrdElementSize(nrrd) instead of nrrdTypeSize[type].  This
meant that situations requiring a bigger allocation were
not always detected as such, causing a later segfault.
Also, nrrdAxesSet_nva(nrrdAxesInfoSize) was not being
called on the newly allocated output.

apply1D.c:
- API CHANGE: in nrrdApply1DLut() and nrrdApply1DRegMap(),
allowing specification of output type (as does 
nrrdApply1DIrregMap())

arrays.c:
- added nrrdTypeMin[], nrrdTypeMax[], nrrdTypeNumberValues[]
- added _nrrdFieldOnePerAxis[NRRD_FIELD_MAX+1]
- reformatted/respelled comments
- allowed dimension field in tables, see read.c changelog

filt.c:
- changed NRRD_AXIS_POS macro usage to NRRD_AXIS_CELL_POS
macro
- nrrdMinMaxClever() renamed to nrrdMinMaxCleverSet()

histogram.c:
- nrrdMinMaxClever() renamed to nrrdMinMaxCleverSet()

map.c:
- changed nrrdHistoEq() to use nrrdApply1DRegMap(), 
changed type of "steady" array to uint, changed type
of non-smart histogram to uint.  One result of this is
that current limitations in nrrdApply1DRegMap() disallow
nin == nout, which also imposes this (new) restriction
on nrrdHistoEq(), alas.
- fixed badness in the map generated by nrrdHistoEq() to
transform values with smartness on: values one bin higher
than a bin ignored via smartness were mapped too low.
- nrrdMinMaxClever() renamed to nrrdMinMaxCleverSet()

measure.c:
- changed NRRD_AXIS_POS macro usage to NRRD_AXIS_CELL_POS
macro

parse.c:
- clarified text of biff error message about failure in
parsing all per-axis fields

private.h:
- added _nrrdFieldOnePerAxis[NRRD_FIELD_MAX+1]

read.c:
- FUNCTIONALITY CHANGE: Previously, "tables" were text files
holding data which was strictly two-dimensional.  When using
"unu lut" and "unu rmap", its often very conveniant to put
the lut/map in a text file, but these are 1D, not 2D.  So,
now tables can represent 1D data.  This has been implemented
by allowing

  # dimension: 1

to be a "comments" preceeding the table data.  The default
dimension for tables is still 2, the line above will
explicitly set it to 1.  But the dimension can not be set to
anything other than 1 and 2, because these are the only
dimensions in which we can imply the axis sizes by the text
layout of the numbers.  Also, "# dimension: 1" can not
appear in the comments *after* a per-axis field, such as
"# spacings: 1 2". 

simple.c:
- FUNCTIONALITY CHANGE: nrrdFitsInFormat() now allows 1D
data to be in a table, see info in read.c changelog.
- clarified comment about action of nrrdHasNonExistSet()

write.c:
- _nrrdWriteNrrd() now takes a "writeData" boolean argument,
which is used by "unu make -h".  This complicates the logic,
but if writeData is zero, you never write data, and the
"data file:" field becomes interesting enough to print,
giving the value of io->dataFN.
- FILE FORMAT CHANGE: this was supposed to have been done
for teem-1.4: normal nrrd magic is "NRRD0001", instead of
"NRRD00.01"
- changes in writing of tables associated with allowing
1D data in tables

test/minmax.c:
- nrrdMinMaxClever() renamed to nrrdMinMaxCleverSet()