Fix PyPI publish flow (#84)

Add on_connection_lost callback to detect socket drops immediately (#42)

Expose an on_connection_lost callback so consumers can react the moment a
socket drops, rather than waiting for the next request to fail. This mirrors
pymodbus's trace_connect(False), firing straight from the asyncio protocol's
connection_lost.

- Add the callback attribute and a safe-invocation helper on AsyncBaseTransport
  (errors raised by the callback are logged, never break transport teardown).
- Wire it into every base transport's _on_connection_lost (TCP, RTU, ASCII,
  RTU-over-TCP).
- Forward it through AsyncSmartTransport to the base transport. Unlike
  on_reconnected, it works even when auto_reconnect is disabled.
- Thread it through the create_async_* client factory helpers.


Claude-Session: https://claude.ai/code/session_01XUgJ5AXGqf5orSNUXtYigk

Co-authored-by: Claude <[email protected]>

Add missing client functions (#41)

* Add missing `read_file_record` and `write_file_record` functions to `AsyncModbusClient`

* Add convenience functions to read/write a single file record

Fix register offset tracking in OrderAwareStruct for sub-register fie…

…lds (#4)

When a register is built from fields smaller than a register (for example
two int8 values packed as ">bbI"), OrderAwareStruct advanced its write
position by the last field's length instead of the accumulated register
length. With word or byte reordering enabled, this spliced the following
multi-register value at the wrong offset, producing corrupted bytes on the
wire. Pack and unpack apply the same transform, so round-trips still
succeeded and hid the issue, but the data sent to a device was wrong.

Advance by the accumulated register length, and add regression tests for
formats that build a register from multiple sub-register tokens.

Fix: make serialization lock per connection (#3)

* Fix smart transport lock ownership

* Expand smart transport state regression test

---------

Co-authored-by: copilot-swe-agent[bot] <[email protected]>

Fix tests

Do reconnect at most once on _must_reconnect

Add missing tests for full coverage

Force reconnect if needed

Add linting to rst docs