VT compliance: automated esctest + visual regression in CI

[deblasis]

Split out from #79 (section 3). Run the VT-compliance suite headlessly in CI.

Status: blocked on CI billing

This fork's GitHub Actions billing is disabled, so there is no runner to host this yet. Filed to track the work; unblocks when CI billing is restored (tracked under the perf/CI umbrella #26).

What already exists (CI-ready)

• esctest harness: windows/scripts/esctest/run-esctest.ps1 + EsctestParse.psm1 (parser/classifier with unit tests).
• Latency probe: windows/scripts/esctest/run-latency-probe.ps1 (proves the transport round-trip).
• Both run unattended and write their own logs (no GUI scrape needed for esctest).

Checklist

• Run esctest headlessly on Windows (ConPTY) per-PR or nightly
• Define pass/fail criteria accounting for the known unanswered-query classes (DECRQCRA etc., see the esctest baseline)
• Visual regression for vttest-style tests (WARP software adapter for headless GPU) -- depends on VT compliance: vttest visual pass on Windows (ConPTY) #507
• Wire into the CI runner
• (Linux baseline is moot per fix(windows): esctest #79 timeouts are missing VT queries, not transport #504: the probe already proved the transport clean)

Context

• Baseline + characterization: feat(windows): esctest VT-compliance baseline over ConPTY (#79) #503, fix(windows): esctest #79 timeouts are missing VT queries, not transport #504, fix(terminal): reply to DECRQM ANSI form and DECXCPR (#79 triage) #505. ConPTY reference doc: docs(windows): add ConPTY sequence-handling reference #506.