[M8] Build runtime-sensitive semantic scenario framework

[Teakowa]

Parent: #9

Depends on: #49

Planning status: provisional until the M7 gate is complete.

Goal

Create repeatable E-level evidence for Workshop behaviors that normalized text comparison cannot prove.

Scope

• Define semantic scenarios with source/input, target/runtime metadata, expected observable behavior, and evidence.
• Prioritize waits, loops, control flow, variable operations, arrays, subroutines, events, player/global state, and type/coercion behavior.
• Support reference-vs-Wright scenario comparison where practical.
• Make scenario results machine-readable and reproducible.
• Add regression scenarios for every runtime-sensitive bug fixed during M8.

Non-goals

• Fully automating the Overwatch client before evidence shows it is required.
• Treating normalized output equality as sufficient semantic evidence.

Acceptance criteria

• High-risk supported semantics have scenario coverage tied to [M8] Finalize v1 compatibility matrix and S/D/N/E release gates #49 release gates.
• Scenario failures identify observable semantic differences rather than only text differences.
• The framework can grow incrementally without hardcoding one test environment.

[Teakowa]

Delivered in commit 1611004 (scenarios/, scripts/run-scenarios.py).

• 7 scenarios cover waits (default/explicit duration), loops (bounded for, hot while), control flow (if/elif/else), variables (initializers, augmented assignment), arrays (literal/append/index/fold), subroutines (def/call), events+state (global/eachPlayer, player variable).
• Each scenario declares target/runtime metadata, expected observable behavior, and machine-checkable evidence; the runner records the emitted Workshop text and analysis findings (target/scenarios-report.json), all 7 pass.
• The environment limitation is explicit: v1 evidence is the compile-time WIR/emission trace + static findings; Overwatch-client execution is out of scope until evidence requires it.