[HazenBabcock]

bin_to_lmchallenge_format.py expands a molecule that was on for several
frames back into one row per frame:

track_length = i3_block['tl'][i]
for j in range(track_length):
    fr = i3_block['fr'][i] + j

That worked against an Insight3 era file, where tracker.c wrote TLEN,
avemlist.c flagged it NOAVERAGE so it survived averaging, and
insight.h defines FRAME as "frame the peak first occured in".

hdf5ToBin(), which replaced that path in Dec 2017, wrote neither field, so
the expansion was silently a no-op:

	in the HDF5	before	after
track_length / tl	1, 3, 10	1, 1, 1	1, 3, 10
frame_number / fr	0, 4, 20	1, 1, 1	1, 5, 21
rows out of lmchallenge	14	3	14

Both halves have to change together, which is why #59 deliberately left tl
alone: each expanded row is given the whole of a, so restoring the track
length on its own would have reported a ten frame track's full photon count
ten times over. a is a track total in the files this package writes, and in
Insight3 era ones too since avemlist flagged AREA as TOTAL, so each row
now gets its share.

Changes

• createFromMultiFit() maps frame_number onto fr and track_length
  onto tl, which is where every other HDF5 field is already translated.
  Insight3 frames are 1-based and HDF5 frames are 0-based, hence the +1.
  Localizations carry neither field, so that path is untouched.
• bin_to_lmchallenge_format.py divides a by the track length.

The converter was executable-only — all module level code reading sys.argv
at import — so nothing could test it. It is now binToLMChallenge() with a
__main__ guard, like every other utility here, and returns the row count.
That accounts for most of its diff.

The new test runs the whole path, HDF5 to .bin to challenge text, and
checks that each track occupies consecutive frames from its own start frame
and that each row carries its per-frame photons. It fails on the parent
commit.

Note that a track with a gap in it is still expanded into consecutive frames.
The .bin format records only where a track started and how long it was, so
the information needed to do better is not there.

---

Full test suite passes, 278 tests.