feat: v0.8.0 — Verification Debt

v0.8.0 — Verification Debt

Closes the original design plan's unfinished verification checks (§3) before the v1.0.0 API freeze turns their answers into contracts — plus a CI-flakiness fix found during real-world gate
validation.

🌊 WebFlux/reactive validation (plan §3 check #4 — never done until now)

• New WireDoctorWebFluxIntegrationTest boots a REACTIVE (Netty, non-servlet) context in CI and asserts reports are written, startup timings populated, and ghost analysis runs
• Reactive entry points added to ghost detection: WebHandler, WebSocketHandler, WebExceptionHandler (joining RouterFunction) — fixes ghost false-positives in reactive apps
• Pinned by test: a RouterFunction bean is never flagged as a ghost
• spring-boot-starter-webflux added test-scope only; README supported-versions note added

🔬 Bean-scope behavior pinned by tests (plan §3 check #2 — doc-only until now)

• New WireDoctorBeanScopeTest asserts the documented behavior per scope:
  • prototype definitions appear as graph nodes but are never instantiated (zero-intrusion promise, now regression-tested)
  • untouched @Lazy beans survive the full analysis untouched
  • FactoryBean product edges resolve to the factory's bean name
  • skipped not-instantiated beans are honestly counted (proxies.notInstantiatedSkipped)
• README Limitation #5 rewritten from vague ("may not fully map out") to these test-backed facts

🤝 Never overwrite a user-set ApplicationStartup (found in plan audit)

• The startup listener previously replaced any ApplicationStartup the app (or another tool) had installed — silently breaking foreign instrumentation
• Now: a pre-existing non-default instance is kept. Buffering ones keep timings working (INFO); foreign ones get a WARN and the timing section degrades gracefully while every other
  analysis runs normally
• getApplicationStartup() verified available on Boot 2.7 → 4.0 (same floor as BufferingApplicationStartup)
• Contract pinned by WireDoctorStartupCoexistenceTest

⏱️ Slow-bean gate jitter margin — wiredoctor.slow-bean-margin-ms (default 20)

• Real-world validation on start.spring.io tripped the gate at 101ms vs a 100ms threshold — pure JVM jitter, not a regression. That exact case is now a permanent test
• A NEW slow bean must exceed threshold + margin to trip; beans inside the margin band stay in the report's slow-bean list but never fail CI
• Mirrors the dual-threshold noise tolerance the startup-time gate already has; 0 restores exact pre-0.8.0 behavior
• Margin surfaces in gates.config and the HTML gate card ((+20ms jitter margin))
• Docs: docs/performance-gates.md + README config section updated

Verified

• 233 tests green (222 autoconfigure + 11 actuator) via full mvn verify — up from 219
• ~60 lines production code, ~350 lines tests — code-light, proof-heavy by design
• Version bumped in all 4 poms; CHANGELOG entry added

Plan §3 checklist after this PR

Check	Status
#1 SPI hook timing	✅ resolved v0.1.0 + CI matrix
#2 Bean scopes	✅ this PR
#3 AOT/native	→ v1.0.0 (deliberate, task 3)
#4 WebFlux	✅ this PR