[HazenBabcock]

Fixes #27.

The issue asked for int32_t instead of int "in many places". The places
that matter turn out to be identifiable rather than a matter of taste, and
there are ten of them, not the 130 that a raw count of int * suggests.

C-to-C calls do not care what sizeof(int) is — the compiler is
self-consistent — and a ctypes.c_int scalar matches a C int at any size.
The assumption only bites where Python hands C a numpy int32 buffer and C
reinterprets it as int[], because ctypes checks the dtype on the Python
side only. A mismatch there is silent memory corruption, not an error.

Scanning every *_c.py argtypes block against the C declarations, Python
passes an int32 array to 24 functions. 14 already used int32_t, all of
them in multi_fit.c and ia_utilities.c — the core was converted at some
point and the rest was not, which is why the tree currently holds two
conventions with no way to tell which is intended.

19199451 converts the remaining ten

file	function	parameters
frc/frc.c	calc_frc	frc_counts
dbscan/dbscan.c	dbscan	db_cat, db_l
	locClSize	cl_counts, cl_id
	recategorize	cl_id, category
L1H/homotopy_general.c	getVisited	vis
L1H/homotopy_imagea_common.c	getPeaks	peak_c
sa_library/grid.c	grid2D	grid, i_x, i_y
	grid3D	grid, i_x, i_y, i_z
sa_library/cs_decon_utilities.c	label	labels
	moments	labels

None has a caller in C, so each change is a declaration and a definition with
no ripple. The other 120 int * parameters never cross the ctypes boundary
and are left alone.

Nothing changes today, and provably so. int32_t is a typedef for int
on every platform this package supports, LP64 and LLP64 alike, so compiling
each file before and after gives byte-identical objects — six checksums are
in the commit message. That identity is also why the compiler cannot check
this work: with int32_t and int the same type, a wrong conversion
produces no diagnostic. The check is the scan.

Note there are two functions named getPeaks. The one taking an int32
array is in L1H. decon_storm/mlem_sparse.c has another whose int *n is an
output count passed as byref(c_int()), so it is not a boundary case, and
that file is not in SConstruct in any event. A scan that keys on function
name must report every definition, not the first, or it hides one behind the
other.

849e4fa3 fixes a function that matched nothing

grid.c declared grid3DZInclusive() and documented it under that name,
then defined grid3DInclusize() — declared-without-definition and
defined-without-declaration at once, wrong since 6a486010. Nothing calls
it. Renamed to match the declaration and the comment, and given the same
int32_t treatment as its two siblings so the file does not keep one
function on the old convention.

Verified by symbol table rather than checksum, since a rename changes the
symbol: exactly one line differs, at the same address, and the disassembly is
identical instruction for instruction once that name is normalized.

---

Full test suite passes, 278 tests. No new compiler warnings.