Expose min_flush_interval on connect() and Client

[mgalletti]

Motivation

The client's write loop coalesces socket writes by enforcing a minimum interval between flushes, hardcoded to 5 ms (_DEFAULT_MIN_FLUSH_INTERVAL). Write-coalescing is the right default when many tasks publish concurrently — one flush amortizes across all their writes.

For sequential publish→await-reply workloads (request/reply, or publish-then-wait-for-ack loops), the interval works against the caller instead: each message pays up to the full 5 ms before its bytes reach the socket, capping a single connection at roughly 200 msgs/s even when the server round trip is well under a millisecond. Today the only way around this is overriding the private _min_flush_interval attribute, which isn't safe across releases.

Changes

• New keyword-only min_flush_interval: float = 0.005 parameter on connect() and Client.__init__. The default is unchanged, so existing callers see no behavior difference.
• 0 is valid and means flush immediately (no coalescing); negative values raise ValueError, validated alongside the existing inbox_prefix checks.
• New read-only Client.min_flush_interval property, matching the style of status / last_error.
• Parameter added at the end of both signatures; docstrings explain the concurrent-vs-sequential trade-off.

Testing

• Three new tests in nats-core/tests/test_client.py: the 5 ms default, connecting with min_flush_interval=0.0 including a publish/subscribe round trip through the write loop, and rejection of negative values.
• Full test_client.py suite passes (178 tests); ruff check, ruff format --check, and codespell are clean