feat: compute models.json in activation via on-the-fly alias merge

[Copilot]

Activation jobs were inlining model alias data (or omitting it entirely), causing code bloat and making model resolution unavailable to runtime scripts. This PR introduces a Compute models activation step that merges pre-computed builtin aliases with per-workflow user overrides at runtime, producing /tmp/gh-aw/models.json in the activation artifact.

Approach

• Builtin aliases live in actions/setup/js/models.json (mirrored from data/model_aliases.json), copied by setup.sh to RUNNER_TEMP — never inlined in compiled YAML
• User overrides (frontmatter models: + imported workflow aliases) are emitted as GH_AW_INFO_MODEL_ALIASES — a compact JSON delta omitted entirely when no overrides are defined
• compute_models.cjs merges the two layers (user wins) → writes /tmp/gh-aw/models.json → logs a collapsible alias table to the step summary
• /tmp/gh-aw/models.json is added to the activation artifact upload

Changes

JS (actions/setup/js/)

• models.json — new pre-computed builtin alias file (auto-copied by setup.sh *.json glob)
• compute_models.cjs — merge helper: loadBuiltinAliases() + loadUserAliases() → mergeAliases() → write + summary
• compute_models.test.cjs — 22 unit tests
• constants.cjs — adds MODELS_JSON_PATH

Go (pkg/workflow/)

• model_aliases.go — MergeUserModelAliases(): like MergeImportedModelAliases but without the builtins layer
• compiler_types.go — UserModelAliases field on WorkflowData (user-level delta only)
• workflow_builder.go — populate UserModelAliases alongside ModelMappings
• compiler_yaml.go — emit GH_AW_INFO_MODEL_ALIASES in generate_aw_info env block (only when non-empty)
• compiler_activation_job_builder.go — generateComputeModelsStep() inserted after generate_aw_info; /tmp/gh-aw/models.json added to artifact paths

Result

For a workflow with no custom models, zero model alias data appears in the compiled YAML. For a workflow with overrides, only the delta is inlined:

- name: Generate agentic run info
  id: generate_aw_info
  env:
    GH_AW_INFO_MODEL_ALIASES: '{"my-alias":["openai/my-model-v1"]}'  # compact delta only
    ...
- name: Compute models
  id: compute_models
  uses: actions/github-script@...
  with:
    script: |
      ...
      const { main } = require('.../compute_models.cjs');
      await main();

[pelikhan]

@copilot use computed models.json when configuring awf

[Copilot]

@copilot use computed models.json when configuring awf

Done in the latest commit. BuildAWFConfigJSON no longer inlines ModelMappings in the static config — instead, BuildAWFCommand appends a Python script to the configFileSetup block that reads /tmp/gh-aw/models.json (downloaded from the activation artifact) and patches apiProxy.models into awf-config.json at runtime, before AWF is invoked. This keeps compiled YAML lean while ensuring AWF always gets the full merged alias map.