Fix existing PET telemetry fields before adding new diagnostics

[eleanorjboyd]

Problem

Recent rollout diagnostics show several existing PET/setup telemetry fields are missing or misclassified, which blocks investigation of Windows rpc_refresh_timeout restarts and setup hangs.

Existing telemetry fixes needed

1. Move numeric fields marked as measurements out of properties and into measures:

   • pet.refresh: envCount, unresolvedCount, workspaceDirCount, searchPathCount, attempt
   • pet.configure: workspaceDirCount, envDirCount, retryCount
   • pet.process_restart: attempt

2. Fix setup.hang_detected so stageDuration arrives alongside duration.

3. Stop classifying PET RPC timeouts as spawn_timeout; use rpc_timeout or method-specific values like rpc_refresh_timeout.

4. Ensure pet.refresh timeout/error events include all known existing context fields: counts, attempt, workspace/search-path counts when available.

5. Improve PET version/build stamping reliability so a single early info timeout does not leave slow/timeout paths as unknown.

Why this matters

Windows PET restarts are mostly rpc_refresh_timeout. Affected sessions are strongly correlated with setup hangs and very slow manager registration, but missing/misclassified fields make the root cause hard to prove.

Acceptance criteria

• setup.hang_detected contains both duration and stageDuration.
• PET refresh RPC timeouts are distinguishable from process spawn timeouts.
• Timeout/error refresh events preserve available context needed for correlation.
• PET version/build/commit is not unnecessarily unknown after recoverable info timeouts.

[copilot-swe-agent]

The agent encountered an error and was unable to start working on this issue: This may be caused by a repository ruleset violation. See granting bypass permissions for the agent, or please contact support if the issue persists. (Request id: D98D:1C9B05:E4FFDC:E8C5AA:6A39F9E8)