Fix Neuralynx unclosed recording header parsing

[Yi-111-a]

Closes #1901

Summary

• tolerate Neuralynx close headers containing the explicit File was not properly closed sentinel
• leave recording_closed unavailable instead of passing the sentinel to dateutil.parser
• add a focused regression test for the malformed close line

Testing

• python -m pytest -q neo/test/rawiotest/test_neuralynxrawio.py::TestNlxHeaderParsing::test_unclosed_recording
• python -m black --check neo/rawio/neuralynxrawio/nlxheader.py neo/test/rawiotest/test_neuralynxrawio.py

[zm711]

This looks fine to me. Let's make sure tests pass and then happy to merge :) Thanks for the contribution.

[zm711]

Failure is currently related to our data repository having some flakiness issues. We will try to get that sorted to ensure this is working with IO tests.

[Yi-111-a]

@zm711 Thanks — noted on the data-repo flake. Closing #1903 as a duplicate of this one (same #1901 goal; this PR is the sentinel-specific fix you already looked at). Happy to wait on the IO tests.

[alejoe91]

@zm711 @h-mayorquin ok to merge this and close #1908 ?

[zm711]

We wanted one update. In #1908 Heberto added a warning. Could we have @Yi-111-a add the warning from 1908 in this one and then we will merge this! Thanks for pinging Alessio!

[Yi-111-a]

Done — pushed b2769e5, adding Heberto's warning from #1908 on top of the existing guard:

if sr:
    dt2 = sr.groupdict()
    if (dt2["date"], dt2["time"]) == ("File", "was"):
        # Cheetah writes this in place of the close time, see #1901
        warnings.warn("Text header does not contain recording closed time. File was not closed properly.")
    else:
        self["recording_closed"] = dateutil.parser.parse(f"{dt2['date']} {dt2['time']}")

The test now wraps the call in assertWarnsRegex and still asserts recording_closed stays unset, so both halves are covered. Verified locally that a properly closed header still parses to 2000-01-01 20:00:00 with no warning emitted — the change only affects the sentinel case.

Happy for this to supersede #1908 — I'll leave that one to @HectorBC and close it if you prefer, since it's your call which of the two lands first.

Re the IO tests: understood, no action needed from me on the data-repo flakiness — the parsing itself no longer routes the sentinel string through dateutil, which is where the ValueError used to come from.

[zm711]

We'll merge this one. I talked to Heberto, so we will close #1908 and keep this one. Let me update and hope that the tests run which would be nice. But we can also get @h-mayorquin to maybe run a test data set against this if he has anything since he has worked on Neuralynx most recently.