Reject NaN deadlines on asyncio to match Trio

[dylanpulver]

NOTE Erasing or replacing the contents of this template will result in your pull
request being summarily closed without consideration!

Changes

Fixes #.

CancelScope on Trio raises ValueError for a NaN deadline; on asyncio the NaN was stored as-is. Every deadline comparison against NaN is false, so the scope never fired — fail_after(math.nan) returned instead of raising TimeoutError, and move_on_after(math.nan) never cancelled. A NaN delay reaching a timeout silently removes the timeout.

This is the same backend inconsistency #1307 fixed for sleep(), on the surfaces that were left. Measured before the change:

call	asyncio	Trio
sleep(nan)	ValueError	ValueError
CancelScope(deadline=nan)	returns	ValueError: deadline must not be NaN
scope.deadline = nan	returns	ValueError: deadline must not be NaN
move_on_at(nan) / fail_at(nan)	returns	ValueError: deadline must not be NaN
move_on_after(nan) / fail_after(nan)	returns	ValueError: deadline must not be NaN

Negative and -inf delays already agree across backends; only NaN diverged. The asyncio CancelScope now rejects it in the constructor and the deadline setter, with Trio's wording, which covers all six entry points since they all funnel through the scope.

Checklist

If this is a user-facing code change, like a bugfix or a new feature, please ensure that
you've fulfilled the following conditions (where applicable):

• You've added tests (in tests/) which would fail without your patch
• You've updated the documentation (in docs/), in case of behavior changes or new
  features
• You've added a new changelog entry (in docs/versionhistory.rst).

Reverting _backends/_asyncio.py fails 18 of the 24 new test cases — every asyncio variant (asyncio, asyncio+uvloop, asyncio+eager) — while the 6 Trio ones still pass, which is the inconsistency itself. Full suite: 3972 passed, 299 skipped, 5 xfailed. pre-commit run clean on both changed files.

No documentation change: fail_after/move_on_after do not currently document delay validation, and sleep()'s :raises ValueError: note from #1307 is the only such mention. Happy to add matching notes if you'd like them.

🤖 Generated with Claude Code

https://claude.ai/code/session_011M5uTyCU4WcNTsPvGrErDo

[agronholm]

Why are we explicitly testing these too? They all come down to CancelScope anyway.