ci: add RustCFML as experimental non-blocking CI target

[bpamiri]

Summary

• Add RustCFML (v0.9.1) to the compat matrix as a non-blocking, experimental engine
• Create standalone test-rustcfml.yml workflow for iterative testing on dev branches
• Include feasibility assessment documenting compatibility analysis

RustCFML is a CFML interpreter written in Rust — a potential lightweight runtime (8 MB / instant startup vs 350 MB / 15s for Lucee). Three of four previously identified blockers (onMissingMethod, cflock, getMetadata) are now implemented. This PR adds visibility into how Wheels behaves on it without affecting CI status.

What's added

Main matrix (tests.yml)

• rustcfml added via include with experimental: true → continue-on-error: true
• Tests SQLite only (RustCFML uses native Rust DB drivers, not JDBC)
• Docker layer caching allowed for the Rust source build
• Added to the matrix summary grid

Standalone workflow (test-rustcfml.yml)

• Triggers on claude/** branches and workflow_dispatch
• Staged: build → health check → smoke test → full suite
• Generates compatibility report committed back to branch
• Every step is continue-on-error: true

Docker infrastructure (tools/docker/rustcfml/)

• Multi-stage Dockerfile: Rust builder → Debian slim runtime
• Custom settings.cfm and app.cfm for SQLite datasource config
• Compose service on port 60100

Feasibility assessment (docs/plans/rustcfml-feasibility-assessment.md)

• Updated evaluation covering v0.4.0 → v0.9.1 progress
• Compatibility analysis against Wheels core features
• Remaining blockers and recommended next steps

Design decisions

• SQLite only: RustCFML uses native Rust database drivers (rusqlite, not JDBC), so CFConfig datasource definitions don't apply. Starting with SQLite minimizes config complexity.
• Separate workflow: The Rust source build takes 15+ minutes. A standalone workflow avoids slowing the main matrix while still providing data on dev branches.
• No --no-cache: Other engines rebuild without cache each run; RustCFML skips this since recompiling Rust from scratch every time is prohibitively slow.
• this.datasources in app.cfm: RustCFML doesn't use CFConfig.json. Datasources are defined via Application.cfc scope using standard CFML this.datasources syntax.

Expected behavior

The RustCFML job will likely fail initially — probably at Application.cfc initialization due to createObject("java", ...) calls or datasource format mismatches. That's the point: each failure maps the compatibility boundary. The experimental: true flag ensures this never blocks PRs or releases.

Test plan

• Verify existing 8 engine matrix jobs are unaffected (no new experimental flag, same continue-on-error: false)
• Verify rustcfml job appears in matrix with continue-on-error: true
• Verify rustcfml failure does not block the overall workflow
• Verify test-rustcfml.yml triggers on claude/** branch push
• Review compatibility report output after first run

https://claude.ai/code/session_01FSAD6qjDXUNiNsBWJncK1N