fix: race condition on connIdIssued in Client.request

[Giulio2002]

Problem

The write to connIdIssued in Client.request() (line 235) on error response (Connection ID mismatch) is not protected by the client mutex. This races with concurrent requestWriter goroutines that read connIdIssued via writeRequest → connect → shouldReconnect, which do hold the lock.

When multiple announce goroutines are spawned by startAnnounce(), they share the same Client and can concurrently enter request(). If two of them receive error responses simultaneously, both write to connIdIssued without synchronization.

Race detected with -race:

Write at 0x00c0f5193528 by goroutine 60999585:
  github.com/anacrolix/torrent/tracker/udp.(*Client).request()
    tracker/udp/client.go:235

Previous write at 0x00c0f5193528 by goroutine 60999531:
  github.com/anacrolix/torrent/tracker/udp.(*Client).request()
    tracker/udp/client.go:235

Fix

Acquire cl.mu before writing connIdIssued = time.Time{} in the error response handler, matching the locking discipline used by writeRequest when reading the same field.

Uses LockCtx(ctx) so the lock attempt is cancelled if the context is already done.

Fixes erigontech/erigon#18901