fix(channel): tear down orphaned WSClient on connect timeout (fixes #197)

[Xuxchloris]

Fixes #197

Problem

When the WebSocket handshake does not complete within connect()'s timeout, the channel rejects with not_connected but leaves the still-handshaking WSClient running in the background:

• it keeps retrying (sockets/timers leak on every timeout),
• its late failures can surface as unhandled exceptions ("WebSocket was closed before the connection was established") that crash the process — the reporter observed 1133 occurrences over 4 weeks and a hard crash during a network blip.

There is also a second, independent unhandled-exception path inside ws-client: aborting a CONNECTING socket (handshake watchdog or close()) calls removeAllListeners() first and then terminate()/close(), and the ws library emits a late error on a terminated CONNECTING socket — with every listener removed that becomes an uncaughtException.

Changes

• channel/channel.ts: on connect timeout, force-close the orphaned WSClient before rejecting, so it cannot leak or keep retrying after the attempt is abandoned.
• channel/types.ts: new connectTimeoutMs option (default 15000, previously hardcoded) so callers can tune the window.
• ws-client/index.ts: keep a no-op error listener after removeAllListeners() in both the handshake watchdog and close(), so aborting a CONNECTING socket stays silent instead of crashing the process.
• channel/__tests__/connect-timeout.test.ts: new tests asserting connect() rejects with not_connected and force-closes the orphaned client, and that a retry starts a fresh client.
• docs (channel.md / channel.zh.md) updated for the new option.

Verification

• npx jest channel — 27 suites, 276 tests passing (including the 2 new tests).
• All changed code paths are unit-tested; no behavior change for successful connects or runtime auto-reconnect (the client is only closed when connect() itself gives up).

[CLAassistant]

All committers have signed the CLA.