[FEATURE] Progressive test coverage gates (strangler-fig ratchet)

[sergio-sisternes-epam]

Is your feature request related to a problem? Please describe.
There is no coverage measurement or gating in CI today. We have no
visibility into coverage trends, and regressions can land silently.

Describe the solution you'd like
Introduce test coverage gates to CI using a strangler-fig ratchet pattern.
Start with observability, then progressively tighten thresholds across
4 phases until we reach steady-state targets of 80% unit / 70% integration.

Baselines (measured locally)

Tier	Baseline	Steady-state target
Unit	80%	80%
Integration	49%	70%

Phase plan

Phase	Unit gate	Integration gate	Intent
1	Report only	Report only	Observability
2	75% fail	Report only	First unit gate
3	80% fail	55% fail	Both tiers gated
4 (steady)	80% fail	70% fail	Steady state

Ratchet rule

After each phase, if actual coverage exceeds the gate by 5+ points,
raise the gate to actual - 3. This prevents regression while allowing
normal fluctuation.

Describe alternatives you've considered

• Hard-gating at 80%/70% from day one -- rejected because the 49%
  integration baseline would block all PRs until gap-fill is done.
• No gates, report only -- rejected because reports alone do not
  prevent regressions.

Additional context
Sub-issues track each phase:

• [FEATURE] Coverage Phase 1: Wire observability into CI #1399 -- Phase 1: Coverage observability
• [FEATURE] Coverage Phase 2: Unit 75% gate #1401 -- Phase 2: Unit 75% gate
• [FEATURE] Coverage Phase 3: Unit 80% / Integration 55% #1402 -- Phase 3: Unit 80% / Integration 55%
• [FEATURE] Coverage Phase 4: Steady-state 80% / 70% #1400 -- Phase 4: Steady-state 80% / 70%