Additional metrics to graph runner

[yushan8]

1. Metrics for graph computation (graphrunner/native.go, graphrunner/BUILD.bazel)

Added a tally.Scope to nativeGraphRunner so it can emit M3 metrics. Three new timers and a gauge are recorded on every Compute call:

• graph_runner.bazel_query_duration — time spent running bazel query
• graph_runner.git_file_hashes_duration — time spent running git ls-tree HEAD to get known source hashes
• graph_runner.target_hash_duration — time spent in targethasher.FromProto (the Merkle DAG hash traversal)
• graph_runner.target_count — number of targets in the computed graph

The scope is inherited from the orchestrator's existing tally scope, so metrics surface as tango.orchestrator.graph_runner.* in M3. NativeGraphRunnerParams gained a Scope tally.Scope field (defaults to noop if nil).

2. Scope wired through orchestrator (orchestrator/native_orchestrator.go)

b.scope is now passed to NativeGraphRunnerParams.Scted to the real tally reporter in production.

3. Hashing optimization (core/targethasher/graph.go

HashRuleCommon was being called twice per rule targ

• Once in toTarget (parallel, result stored in target.HashWithoutDeps)
• Once again in HashRecursively's default case (red

Fix: removed target.HashWithoutDeps = []byte{} fromget.Hash needs to be the sentinel), thenHashRecursively reads target.HashWithoutDeps directly instead of re-running HashRuleCommon. A fallback still calls HashRuleCommon if HashWithoutDeps is somehow empty.