[Domain] model hazard detection delay before reaction

[SilverSupplier]

Summary

• Model a hazard detection delay distinct from the existing reaction delay (report section 4.3). An agent now senses a hazard on entering its range, detects it only after Agent::detectionDelaySeconds, and then becomes aware after reactionDelaySeconds.
• The environment reaction state tracks when sensing began (hazardSensedSinceSeconds); with a zero detection delay the behavior is identical to before, so existing hazard tests are unaffected.
• Populate detectionDelaySeconds per agent at spawn with a deterministic, seed-derived value (independent beta22 draw, so existing random draws are unchanged).
• Add ScenarioEnvironmentHazardSystem_DelaysDetectionUntilDetectionDelayElapses covering sensed-but-not-detected and detected-then-aware.

Area

• Engine
• Domain
• Application

Architecture Check

• I kept the dependency direction application -> domain -> engine.
• I did not add Qt UI code to src/domain.
• I did not add domain or application dependencies to src/engine.
• I used src/ as the include root.

Verification

• cmake --preset windows-debug
• cmake --build --preset build-debug
• ctest --preset test-debug
• Not run (reason below)

Ran the release / no-app equivalents: cmake --build --preset build-no-app-release, ctest --preset test-no-app-release -> 100% pass (includes the new detection-delay test). Full Qt app build (cmake --build --preset build-release --target safecrowd_app) also succeeds.

Risks / Follow-up

• With a zero detection delay the detect/react pipeline is equivalent to the previous immediate-detection behavior, so existing hazard reaction tests keep passing.
• reactionDelaySeconds remains unpopulated at spawn (a pre-existing condition); only detectionDelaySeconds is now seeded.