[TheNetherWatcher]

Closes #597.

The skip added in #596 is only correct while two things stay true:

• _update_noise doesn't change qrs_thr.
• _update_qrs sets last_qrs_peak_num to the current peak.

This PR:

• Uses the comment suggested in Document what the XQRS backsearch skip depends on #597 above the skip in _backsearch.
• Adds test_xqrs_backsearch_skip_assumptions, which runs XQRS on record 100 with a 2-minute low-amplitude gap followed by scaled-down beats (so backsearch actually finds QRS complexes) and checks both properties after every update.

I first tried a test comparing the skip against a full rescan, but it still passed when I broke either property, because the data never hit the case where they matter. The direct check fails for both changes:

• _update_noise also recalculating qrs_thr → fails
• _update_qrs setting last_qrs_peak_num = peak_num → fails

No change in behaviour, so no changelog entry. ruff check is clean and pytest -m "not network" passes (83 tests).

[tompollard]

Thanks @TheNetherWatcher, this looks good! Checking the two assumptions directly makes sense.

One small optional addition: the comment says the result depends on qrs_thr and last_qrs_ind, but noise_spy only checks qrs_thr.
Checking both would cover the comment fully:

def noise_spy(peak_num):
    qrs_thr, last_qrs_ind = xqrs.qrs_thr, xqrs.last_qrs_ind
    update_noise(peak_num)
    assert (xqrs.qrs_thr, xqrs.last_qrs_ind) == (qrs_thr, last_qrs_ind)

Happy to merge as it is, or with that change.

[TheNetherWatcher]

Thanks! Good point, I've added it. The noise check now compares (qrs_thr, last_qrs_ind), and it fails if _update_noise changes last_qrs_ind.

[tompollard]

Looks good to me, thanks @TheNetherWatcher!