feat(observability): collect host scheduler-pressure metrics (PSI, procs, reclaim, per-NUMA)

[nv-yna]

Default Tachometer host capture lacked runnable/blocked process counts, host context-switch counters, major faults/reclaim, PSI, and NUMA memory breakdowns. This change enables those node-exporter collectors on worker nodes and keeps each NUMA node distinct in the raw capture.

Behavior

• Add stat, vmstat, pressure, and meminfo_numa alongside the existing cpu, infiniband, meminfo, and processes collectors. The vmstat field filter includes pgsteal_* while retaining major faults. Explicit recipe commands continue to take precedence.
• Preserve the exporter’s NUMA node label as numa_node in raw metric names, separately from host metadata. This applies to NUMA memory gauges and allocation counters, and retains feat(observability): scrape a process exporter with tachometer by default #413’s process-state and passthrough fixes.
• Add cumulative PSI stall totals, in microseconds, to the existing optional local host sampler. It remains gated by observability.enabled, covers only the sweep/orchestrator host, and has no dependency on the closed all-node sampler PR feat(observability): host sampler on every node + scheduler-contention metrics #356.
• Missing or unreadable PSI files produce unavailable data, not zero pressure. This does not extend node-exporter placement to dedicated frontend/client nodes. Collector overhead has not been measured for this change.

Deployment

Use make tachometer-scraper to build a scraper containing this change, or use a release that includes it. Older scraper binaries can collapse NUMA labels even when the exporter collects them.

Validation

• Rebased onto 0ecc0bdb8f8fc507d769cda0449aec70c52286de after feat(observability): scrape a process exporter with tachometer by default #413 merged; preserved the combined collector set.
• make check: 2,152 passed, 2 skipped, 6 deselected. Final source/test lint and formatting passed.
• cargo test --locked -p tachometer-scraper: all 34 tests passed; Rust formatting passed.
• A regression reproduces NUMA label loss before the fix. Tests cover separate NUMA nodes for MemFree and allocation counters, unchanged host metadata and ordinary memory names, actual launched collector arguments, reclaim matching, custom-command precedence, and PSI present/absent cases.
• No new cluster workload was launched. Hosted CI passed all seven checks, including the full Rust workspace tests and release build, on 55bb240d2bb379ac9667db08ae7633d046dadece before merge.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 95.00000% with 1 line in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (main@0ecc0bd). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
src/srtctl/analysis/host_sampler.py	93.33%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #415   +/-   ##
=======================================
  Coverage        ?   75.25%           
=======================================
  Files           ?      109           
  Lines           ?    16115           
  Branches        ?        0           
=======================================
  Hits            ?    12127           
  Misses          ?     3988           
  Partials        ?        0           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.