log: support discard action on write timeout

[bb7133]

What problem does this PR solve?

Config.Timeout currently always panics when a log write/sync waits too long for the writer lock. TiDB needs a configurable behavior so operators can choose to discard subsequent logs instead of panicking when log I/O is stuck.

What changed and how does it work?

• Add Config.TimeoutAction with valid actions panic and discard. Empty action keeps the existing panic behavior.
• Keep LockWithTimeout as the panic-compatible wrapper.
• Add LockWithTimeoutAction for callers that want discard behavior.
• In discard mode, timed-out Write returns len(bs), nil and timed-out Sync returns nil. After the first timeout, subsequent calls discard immediately while the writer lock is still unavailable, instead of waiting the full timeout repeatedly.

Check List

Tests

• Unit test

go test ./...

[YangKeao]

LGTM