Close native update coroutines cancelled while waiting for processor capacity

When BaseUpdateProcessor.process_update is cancelled while waiting for its concurrency semaphore, the supplied native coroutine never reaches do_process_update and remains unclosed. Close it during cleanup if admission has not occurred. Once admitted, the subclass retains ownership; Futures and arbitrary awaitables are left untouched.

Adds native processor regressions for the queued native coroutine, untouched Future/custom awaitable (including a close method), slot recovery, and admitted subclass ownership. The regression fails on master with CORO_CREATED instead of CORO_CLOSED and passes with this change.

Related: #5390

Validation on master 9a38d8f516fd14c56ef51dc21817d8fbe44c267c, Python 3.14.9, locked upstream test/lint dependencies:

• python -B -m pytest --disable-socket tests/ext/test_baseupdateprocessor.py: 13 passed (before the source change: 1 failed, 12 passed).
• ruff check src/telegram/ext/_baseupdateprocessor.py tests/ext/test_baseupdateprocessor.py: passed.
• ruff format --check src/telegram/ext/_baseupdateprocessor.py tests/ext/test_baseupdateprocessor.py: passed.

These are offline processor tests; no Bot, transport or account is constructed. The complete repository suite and other Python versions were not run. Cancellation before process_update starts executing remains the caller's responsibility.