fix(server): detect dead clients and reclaim their sockets

[tentious]

Problem

Fixes #470. The server only wrote to control channels, so a half-closed client could leave its handle, data-channel pool, and service listener alive indefinitely.

Changes

• read control channels concurrently with heartbeat and data-channel writes
• remove a finished channel by its unique session key so reconnects are not disturbed
• use a weak map reference so cleanup tasks cannot retain server state during shutdown
• reject unexpected client data as a protocol violation
• retarget and rebase the contributor change onto dev

Regression coverage

• half-closed control channel removes both map indexes and releases the listener with heartbeat disabled
• dropping the channel map releases the pool without a reference cycle
• end-to-end client shutdown releases the service listener while the server remains running

Validation

• format and docs
• clippy and unit tests with default and rustls CI features
• all 134 supported feature combinations
• native-tls and rustls integration suites (8/8 each)

[sjtrny]

Thanks!