Fix SimpleQueue get and argument handling - #9641

Summary

• Route SimpleQueue.get(block=False) through the existing nonblocking removal path. Previously a threaded nonblocking get removed an item without consuming its semaphore permit, so later consumers could observe stale permits.
• Publish each enqueued item and its permit before releasing the buffer lock, preventing another thread from observing an item before a nonblocking getter can claim its permit. Construct an empty-queue exception after releasing the buffer lock.
• Apply Python truth testing to the block argument for put and get, after argument binding; preserve explicit None as false.
• Ignore the timeout for nonblocking calls. For blocking calls, convert __index__ before __float__, validate signed nanosecond bounds and round up as CPython does, reject NaN/negative values appropriately, and use checked deadline addition. Infinite or very large timeouts no longer reach a panicking Duration::from_secs_f64 path.

Scope

Generic argument-binding diagnostics, lower-level numeric-slot diagnostics, existing item ownership, waiting/polling behavior and direct-handoff/fairness differences are not addressed. put still acquires the same two mutexes but now holds them together briefly, so contention effects are unbenchmarked. If a platform's Instant cannot represent a valid deadline, this returns OverflowError; CPython instead saturates its internal deadline.

Validation

• Traced semaphore and buffer states across mixed get(False), get(), get_nowait(), put and qsize calls, including lock ordering and stop-the-world interactions. Existing Lib/test/test_queue.py exercises mixed and concurrent nonblocking calls.
• Compared CPython 3.15.0 queue and time conversion source and reproduced truth testing, ignored nonblocking timeouts, __index__ precedence, propagated callback errors, nanosecond rounding, NaN and overflow cases.
• Independent source review for all four commits and normal commit hooks passed.
• RustPython was not built or executed locally; stale-permit and panic paths are inferred from source, not locally reproduced. No tests were added or changed.