[Jack-Coutts]

Fixes #187.
Re-applies #217 onto main.

The bug

Changing the mass tolerance re-extracted every EIC at the default 0.2 min tR window. A window set with Apply window was lost, and so was the wider window Add Compound extracts at from a compound's tR tolerance. Plots narrowed and the tR Window field changed. A window narrower than 0.2 was widened instead.

The fix

regenerate_all_eics_with_mass_tolerance in eic_importer.py now reads each sample's stored EIC windows and re-extracts each compound at its stored window. It still widens a window that would not cover the integration offsets. A compound with no stored EIC in a sample keeps the old rule, the default window or the offset minimum. The shared window passed to the extraction is now 0.0, so it no longer widens a narrower stored window.

The user guide and the mass tolerance reference say that a tolerance change keeps each EIC's tR window.

Checks

• tests/test_mass_tolerance_keeps_tr_window.py drives the window. It writes a CDF, imports it, sets Glucose's window with Apply window (0.9, and 0.15 in a second case), adds Lactate with a 0.6 tR tolerance through Add Compound, then changes the mass tolerance. It asserts each stored window and the EIC's time range. Without the fix both cases fail, for example {'Glucose': (0.2, 4.8, 5.2)} != {'Glucose': (0.9, 4.1, 5.9)} and {'Lactate': (0.2, 4.8, 5.2)} != {'Lactate': (0.6, 4.4, 5.6)}. With it, 2 passed.
• uv run python -m pytest -q: 673 passed.
• bench/reference.py check: exported results unchanged.

Left out of #217

Its test in test_rt_window_centering.py, which stubbed the CDF reader and called the function directly. The new test runs through the window in its own file. The subquery and second bind parameter in the per-sample query are replaced by a separate query for the stored windows, so this does not touch the query's WHERE clause or parameters.

[Jack-Coutts]

Review verdict for head 4de1cca: CLEAN. Reviewed by GPT-6.1 Sol.

What Sol ran:

• The full suite: 673 passed.
• The new tests: 2 failed with src restored from origin/main, then 2 passed on the head.
• The results check: zero changed cells against the reference.
• Two tolerance reloads in a row through the window, with applied windows of 0.9 and 0.15. Windows, time ranges and exported raw and corrected areas held. A window narrower than its offsets need widened once to 0.5 and stayed there, so the widening does not build up.
• The shared 0.0 passed to _extract_all_eics_for_file. It is used only in the max with each target's window, so nothing downstream sees it.
• Soft-deleting a compound. Only compounds.deleted changes, so its EIC rows keep deleted = 0, and fix(eic): regenerate deleted compounds and samples on a mass tolerance change #306's rebuild of deleted compounds keeps their stored windows.

What it found:

• No blocker or warning.
• main allows duplicate live EIC rows for one sample and compound if import_eics runs twice. The window then wins in no set order. The UI prevents a second folder import, so this is not this pull request's to fix.

Nothing was fixed in review.