Make Windows process diagnostics resilient to CIM timeouts

[badcuban]

Priority: P1

Why

Windows process diagnostics can time out and become a Diagnostics failure. The current Windows query shells out to PowerShell with a 1 second timeout and performs per-process CIM lookups, which is fragile under load.

Diagnostics evidence

From C:\Users\Will\.t3\userdata\logs\server.trace.ndjson* on May 26, 2026:

• Recent failures included runProcess with ProcessDiagnosticsError: Failed to query process diagnostics. timed out.
• The failures came from the process diagnostics path used by Settings -> Diagnostics.
• apps/server/src/diagnostics/ProcessDiagnostics.ts sets PROCESS_QUERY_TIMEOUT_MS = 1_000 and readWindowsProcessRows() runs Get-CimInstance Win32_Process plus a per-process Get-CimInstance Win32_PerfFormattedData_PerfProc_Process lookup.

Acceptance criteria

• Windows process diagnostics avoids per-process CIM calls or performs them in a batched/cheaper way.
• The diagnostics page returns partial or cached process data instead of turning the whole read into a trace failure when CPU details are slow.
• Timeout behavior is tuned for Windows machines under load.
• Add Windows-focused coverage for timeout/partial-result behavior.

[badcuban]

Completed in the 0.0.3-0.0.4 diagnostics work. Windows process diagnostics now uses one bulk process/perf query, has a longer timeout, filters the diagnostics query process, and returns a diagnostics result with an error field instead of failing the whole read. Covered by ProcessDiagnostics tests.