Preserve float-subclass values in time.sleep conversion - #9292

Summary

Read the stored value of float instances and subclasses in time.sleep instead of calling an overridden float. This follows CPython 3.15.0 PyFloat_AsDouble behavior. Keep index precedence and the existing non-float conversion path, rounding, range checks, audit ordering and platform sleep implementation unchanged.

A float subclass can currently change whether sleep rejects a negative value by overriding float, or raise an exception that CPython never requests. This narrow native fix extracts that behavior from #8954 without adding an unused PyTime helper module or moving the surrounding conversion code. No Lib changes or new Rust Python-behavior tests.

Validation

Tested base: 43368209c505c8807f5efdf3a39729bb30866722.

• Source-bound default-features + capi CLI build.
• Eight bounded external observations using only zero/negative durations: baseline matches CPython in four, candidate matches all eight, including conversion call counts and exception identity.
• One unchanged AST-selected canonical method, test.test_time.TimeTestCase.test_sleep_exceptions, passes on baseline, candidate and CPython 3.15.0. Its AST matches the pinned canonical source. The class setUp assignment self.t = time.time() was omitted because this method does not use it; the method and assertions are unchanged. This is selected-method evidence, not whole-module coverage.
• VM threaded/non-threaded checks, strict all-target VM Clippy and changed-file hooks passed.

Full workspace/CAPI runtime suites, release test_time, the whole time test module and positive-duration sleep tests were not run locally. The external observations do not establish cross-platform scheduling behavior or broad PyTime parity.

AI assistance

AI assistance was used for source analysis, implementation and validation preparation.