Clean up logging and remove double-logged errors

[yushan8]

Removed double-logged errors (logged then returned — caller logs again):

• core/workspace/gitrequest.go — 5 Errorw calls in Apply() (fetch, diff, patch, commit, submodule). The orchestrator already logs when ApplyRequests fails.
• core/bazel/bazel.go — Errorw for bazel executable detection. Caller logs.
• core/bazel/query.go — Errorf in wrapQueryFailure. The error message already carries stderr; logging it separately was redundant.
• orchestrator/native_orchestrator.go — 11 Errorw calls across the entire GetTargetGraph method. All errors are returned as ClassifiedError and the controller boundary emits failure metrics + logs.

Corrected log levels:

• controller/getchangedtargets.go:55 — validation failure Error → removed (linter removed the log entirely since it's returned with classification)
• controller/getchangedtargets.go:293 — background goroutine cache-write skip Error → Warn (not a service failure, can't return)
• controller/gettargetgraph.go:107 — treehash cache miss Info → Debug (normal hot-path noise)
• core/bazel/bazel.go:94 — constructor detail Info → Debugw
• core/bazel/query.go:40 — bazel query detail Infow → Debugw

Fixed unstructured logging:

• orchestrator/native_orchestrator.go:137 — Errorf → Errorw with zap.Error()
• core/bazel/query.go:103 — Debugf → Debugw with zap.Int()

[sbalabanov]

in many cases it is beneficial to wrap the error with fmt.Errorf to include additional metadata and preserve call stack in the error chain.