feat: add durable producer flush

[infiniteregrets]

Summary

• add Producer.flush() as a durable, non-terminal boundary for prior submissions
• force the current partial batch without waiting for linger, then wait for every prior record acknowledgement
• keep later submissions outside the boundary and preserve producer reuse
• make terminal submit/ack failures sticky across tickets, submit(), flush(), and close()
• keep an ordered flush running if its caller is cancelled
• document the new checkpoint in the centralized producer example

Why

Previously, closing the producer was the only way to force a partial batch and wait for durability. Applications need an explicit durability checkpoint while continuing to use the same producer.

This implements the Python SDK portion of s2-streamstore/s2#666.

Validation

• uv run --group test pytest tests/test_producer.py -q — 4 passed
• non-integration test suite — 82 passed, 166 deselected
• Ruff lint and format checks
• mypy
• Sphinx build with warnings treated as errors