Tracer: observed-first replay_case v2, bundle export/runtime, CLI, and remove generated fixtures

[AlexanderOnischenko]

Motivation

• Migrate tracer events to an observed-first model where runtime logs carry only observed/observed_error and dependencies for deterministic replay bundles rather than embedding expected outcomes.
• Provide self-contained, portable case bundles and a compact runtime API to load/run those bundles for deterministic testing and local regeneration of fixtures.
• Exclude large auto-generated case bundle fixtures from the repository and replace them with placeholders so bundles are regenerated locally as needed.

Description

• Replace the old replay_point API with log_replay_case(...) (v2) that validates inputs, enforces XOR of observed/observed_error, and requires structured requires entries in src/fetchgraph/replay/log.py.
• Rework exporter logic in src/fetchgraph/replay/export.py to select replay_case events, produce .case.json bundle files via write_case_bundle, resolve requires with collect_requires, copy resource files with copy_resource_files, and add multi-export helper export_replay_case_bundles.
• Add runtime helpers load_case_bundle and run_case and ReplayContext enhancements in src/fetchgraph/replay/runtime.py to load bundles and resolve resource paths for handlers.
• Introduce a small tracer facade package (src/fetchgraph/tracer/*) with handler hub, CLI (fetchgraph-tracer entry point), validator validate_plan_normalize_spec_v1, and Makefile/pyproject.toml wiring (fetchgraph-tracer script and tracer-export Make target).
• Update the user-facing CLI in src/fetchgraph/cli.py and Makefile to target tests/fixtures/replay_cases and the new export-case-bundle semantics.
• Update tests in tests/test_replay_fixtures.py to use .case.json bundles: fixed cases assert exact expected equality and known_bad cases validate that the handler output fails the validator.
• Remove large generated bundle fixtures and resource copies from tests/fixtures/replay_cases and add .gitkeep placeholders plus tests/fixtures/replay_cases/README.md documenting local regeneration.

Testing

• No automated test run was executed as part of this change; pytest was not invoked in the rollout.
• The test harness was updated: tests/test_replay_fixtures.py now targets tests/fixtures/replay_cases/*.case.json and distinguishes fixed vs known_bad flows but requires regenerated bundles to be present for full verification.
• Fixture deletions were committed so repository contains placeholders and guidance (tests/fixtures/replay_cases/.gitkeep and README.md), and test runs will be skipped until bundles are recreated locally.

---

Codex Task

[AlexanderOnischenko]

@codex, please review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9d66581448

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Make overwrite actually replace existing targets

When fixture-demote runs with --overwrite, it skips the existence checks but still calls _GitOps.move, which uses git mv without -f. git mv -h states -f, --force is required to move/rename even if the target exists, so in the default git_mode=auto a tracked destination fixture will cause the demotion to fail exactly when overwrite is requested. Consider deleting the destination before moving or using git mv -f when overwrite is set.

Useful? React with 👍 / 👎.