Fix trim expansion loops testing the already-included cell

[lispandfound]

Fixes #86

Problem

The two expansion loops in trim_array_to_target_length tested a cell that is already inside the window rather than the candidate cell about to be absorbed:

# before
while left > 0 and slip_function[left] >= keep_threshold:            # left is already included
    left -= 1
while right < len(slip_function) and slip_function[right - 1] >= keep_threshold:  # right-1 already included
    right += 1

So whenever a boundary cell was above keep_threshold, the window swallowed its neighbour without ever checking that neighbour — typically a sub-threshold cell the shrink loop had just removed.

Measured impact

input	target	before	after	tolerance (±2*dx)
[1,1,9,1,1]	1	(0,4) len 4	(0,3) len 3	before violates (|4-1|=3 > 2)
[1,1,1,9,1,1,1]	1	(1,5) len 4	(1,4) len 3	before violates (|4-1|=3 > 2)
[0,0,9,0,0]	1	(2,3)	(2,3)	unchanged

Trace for the first case: the shrink loop correctly reaches (0, 3), then the second expansion loop sees slip_function[right - 1] == slip_function[2] == 9.0 >= 3.0 and grows right to 4, absorbing cell 3 whose value is 1.0 — below the threshold and never examined.

Correction to the issue as filed

Issue #86 cited [1,1,9,1,1] with target_length=3 returning (0, 5) — "the whole array" — as evidence. That result is actually within the documented target_length ± 2*dx tolerance (|5-3| = 2), so it is not by itself a defect; I have noted that on the issue. The tolerance violation at target_length=1 above is the real evidence, and it is what the new test asserts.

Cases where the absorbed cell is exactly zero are also not observable, because the zero-stripping loops further down clean them up. The bug only shows with sub-threshold but nonzero neighbours.

Changes

• source_modelling/trim.py — test slip_function[left - 1] and slip_function[right], the cells actually being considered.
• tests/test_trim.py — add test_trim_expansion_does_not_absorb_sub_threshold_cells, asserting both the exact window and the documented tolerance.

Verification

• 2 of the 3 new parameters fail without the fix, all pass with it.
• pytest tests/test_trim.py — 12 passed.
• ruff check, ruff format, numpydoc lint clean.

Note this touches the same function as #88 (the bounds-guard ordering a few lines below), so those two branches will want a trivial rebase depending on merge order.

🤖 Generated with Claude Code