fix: restore quiet retry for interrupted Lavish polls

[withally]

Summary

• restore the Lavish adapter poll subcommand from squash 3a16740 as overlay 2 of 3
• retry only the exact two-line interrupted poll SERVER_ERROR, with 12 retries and a 5 second default delay
• keep the generic process-event runner adapter-agnostic
• preserve ordinary capture and wake behavior for feedback, ended or missing sessions, unrelated errors, and bounded exhaustion

Overlay ledger

• Position: overlay 2 of 3; overlay 1 spawn-isolation remains untouched; no overlay 3 is added
• Why retained: live board-listen pings still emitted Lavish Editor poll response was interrupted / SERVER_ERROR after lavish-axi 0.1.55, waking the fleet immediately after the upstream snapshot
• Retirement: remove this overlay when Lavish swallows this exact transient interruption upstream or the Firstmate Lavish adapter lands upstream
• Upstream issue: Lavish process-event poll interruption wakes fleet instead of retrying quietly kunchenguid/firstmate#2773

Verification

• tests/fm-procevent.test.sh - all procevent tests passed
  • exact retried interrupts create no captured result and no check wake
  • feedback after retries captures and wakes
  • exhaustion captures and wakes after 12 retries, 13 attempts total
  • unrelated SERVER_ERROR is captured and wakes without retry
• bin/fm-doc-audience-check.sh - ok, 71 surfaces and 260 local links
• bin/fm-lint.sh - shell lint phase passed; workflow lint could not run because pinned actionlint 1.7.12 is not installed locally
• git diff --check origin/main - clean
• git diff --stat origin/main - only adapter, tests, and process-event-sources pointer; no AGENTS.md or lavish-axi product changes

Direct PR. No no-mistakes run. Do not merge without explicit authority.