Instrument fork admission timing

[sjmiller609]

Summary

Adds targeted tracing around fork admission so parallel fanout traces can show where time is spent before restore:

• traces metadata exact-name scans with file/scanned counts and static caller labels
• traces fork metadata save, lock wait, and lock hold time
• traces snapshot memory readiness and guest directory clone work for fork/snapshot paths
• applies the metadata name-exists tracing to snapshot fork name checks too

The spans intentionally avoid recording requested instance names.

Testing

• go test ./lib/instances -run 'TestForkInstanceFromStandby|TestForkInstanceFromStopped|TestGetInstance|TestListInstances|TestSnapshot' -count=1\n- go test ./lib/instances -run 'TestSaveForkMetadataSerializesNameAdmission|TestForkInstanceRejectsDuplicateNameForNonNetworkedSource|TestForkInstanceFromStandbyCancelsCompressionJobAndCopiesRawMemory|TestForkSnapshotFromCompressedSourceCopiesRawMemory' -count=1\n- go test ./lib/network -count=1\n\nNote: I did not run the full go test ./lib/instances package to completion locally; it entered longer integration-style tests, so I used the focused fork/query coverage above for this instrumentation-only patch.