fix: prevent crash when disconnecting etcd connection

[datlechin]

Summary

Fixes a crash (EXC_CRASH SIGABRT) when disconnecting an etcd connection while requests are in-flight (ping, watch, or auth).

Root cause: EtcdHttpClient.disconnect() calls session.invalidateAndCancel(), but in-flight async requests had already captured the URLSession reference outside the lock. When the continuation closure later calls session.dataTask(with:) on the invalidated session, CFNetwork throws an Objective-C NSException — which is uncatchable by Swift do/catch — causing abort().

Fix: Added a sessionGeneration counter that increments on disconnect. All three dataTask creation sites (performRequest, watch, authenticate) now re-acquire the lock and verify the generation matches before calling dataTask. If the session was invalidated, they throw EtcdError.notConnected instead of crashing.

Crash sites fixed (all in EtcdHttpClient.swift):

• performRequest — line 717 (general API calls, ping)
• watch — line 574 (key watch long-poll)
• authenticate — line 817 (token refresh)

Test plan

• Connect to an etcd server
• While connected, disconnect — verify no crash
• While a watch is active, disconnect — verify no crash
• Rapidly connect/disconnect — verify no crash
• Verify normal etcd operations still work after reconnecting