Language cleanup: two workstreams (WS01 mechanical, WS02 data/outcomes) for architecture review

[handcaught]

Summary

Scopes the Terraform-shaping work the VSCode extension surfaced as two workstreams for architecture review before implementation begins. No code changes yet — just the workstream plans.

• WS01 — Mechanical schema cleanup. Low-risk, all touch workflow/schema.go: workflow { name = ... } reshape with nested policy, type = string type expressions (drop the hand-rolled parseVariableType / TypeToString), outcome "default" {} replaces the default_outcome attribute, environment refs as traversals (environment = environment.shell.ci), register cty/function/stdlib for the full Terraform-equivalent function set (~50 funcs incl. startswith/substr/replace/format/etc.), VSCode grammar update. Hand-rolled duplicates (jsonencode, base64encode, etc.) are dropped in favor of stdlib — no override fallback.

• WS02 — data block + outcome semantics. Higher-risk, touches the engine runtime: outcome next becomes a node traversal (step.foo, state.done) with bare return/continue keywords replacing the magic-string sentinels; shared_variable is replaced by data "internal" "name" (extensible block, ready for future remote kinds); shared_writes = { ... } becomes per-target write { target = ..., value = ... } blocks inside outcomes (singular write since each updates one target — data.x.y.value = expr can't be an HCL attribute LHS); engine runtime store renamed SharedVarStore → DataStore; VSCode grammar update.

Migration strategy is hard break with helpful errors — same pattern as parse_legacy_reject.go. No dual-support window.

WS01 lands first to absorb mechanical schema churn; WS02 then lands on a clean base.

Branch targets main; once the language-cleanup phase closes, main merges into adapter-v2. Workstream files live in workstreams/language_cleanup/ following the adapter_v2/ precedent.

Test plan

• Architect review of workstreams/language_cleanup/README.md for phase framing.
• Architect review of WS01-mechanical-schema-cleanup.md — confirm scope split is correct; flag any concerns about the default_outcome → outcome "default" {} shape (it's the one item where WS01 changes semantic surface, not just syntax).
• Architect review of WS02-data-and-outcome-semantics.md — confirm the data "<kind>" "<name>" shape (with only internal supported initially) is the right extension point, and confirm the per-target write { target = ..., value = ... } block shape is acceptable given the HCL grammar constraint that prevents data.x.y.value = expr attribute LHSs.
• Sign-off on hard-break (no dual-support) migration strategy.