feat(capture): add S1 app parser registry with golden fixture tests

[heming-gmh]

Summary

Introduce a lightweight, baseline-plus-patches parser registry for the S1 capture enrichment layer, so the project can grow from one monolithic module (s1_parser.py) into a family of small, independently-testable app parsers — without changing a single byte of downstream behaviour.

Why

The current s1_parser.enrich(capture) merges three responsibilities in one module: generic AX→S1 extraction, browser URL extraction, and (on another branch) editor/terminal metadata. This works for a small surface area, but it does not scale as the project adds parsers for Slack, Linear, Office, Figma, Notion, terminals, and other high-value contexts.

This PR splits infrastructure from implementation:

The Baseline + Patches Model

1. enrich() always computes a generic baseline (focused element, visible text, url=None).
2. Registered AppParser implementations then run in priority order.
3. Each parser may return an S1Patch that selectively overrides fields.
4. Later parsers can compose on top of earlier ones — e.g. a Linear parser can match linear.app in the URL that the browser parser already extracted.

This is intentionally not a first-match-wins dispatch. Composition lets an app that is both a browser AND a special-purpose tool (e.g. a browser-based IDE) get both URL extraction AND app-specific metadata.

Benefits

• Reviewable parsers: golden fixture tests (input.json → expected.json) let maintainers inspect one raw AX input and one expected S1 output for each parser PR.
• Fault isolation: a broken parser never takes down the whole capture pipeline — exceptions are logged and the baseline fields still ship.
• Zero downstream churn: focused_element, visible_text, and url are unchanged; timeline, FTS indexing, and dedup hashing all work without modification.
• Low ceremony for contributors: a future parser author only needs to implement AppParser (4 members), register it, add a fixture directory, and open a PR.

What Changed

New files

File	Purpose
app_parsers/__init__.py	Parser registry — register(), apply_parsers(), builtin auto-registration
app_parsers/base.py	FocusedElement, ParseContext, S1Fields, S1Patch, AppParser Protocol
app_parsers/browser.py	BrowserParser — migrated from s1_parser._extract_url, behaviour identical
tests/test_s1_registry.py	5 unit tests: priority ordering, patch composition, exception resilience, app_context write semantics
tests/test_s1_golden.py	Parametrised golden fixture runner (4 fixtures)
tests/test_s1_integration.py	4 integration tests: scheduler+enrich+FTS pipeline, timeline _format_events rendering
tests/fixtures/s1/*/	4 fixture directories (chrome_url, safari_bare_domain, generic_cursor_textarea, non_browser_url_textfield)

Modified files

File	Change
s1_parser.py	Remove _BROWSER_BUNDLES, _extract_url; rewrite enrich() as baseline + apply_parsers(); re-export FocusedElement from base.py
tests/conftest.py	Add autouse fixture to reset parser registry before each test

Not in this PR

• No Slack, Linear, Office, editor, or terminal parser behaviour.
• No dynamic plugin loading or entry-point discovery.
• No schema_version change.
• No AppleScript, browser automation, OCR, or new permissions.

Test Plan

• 9 existing test_s1_parser.py tests pass unchanged
• 5 registry unit tests cover ordering, composition, error isolation, app_context semantics
• 4 golden fixture tests cover browser + non-browser URL extraction scenarios
• 4 integration tests cover scheduler→enrich→FTS and timeline _format_events contracts
• Full test suite: 82/82 pass

Backwards Compatibility

enrich(capture) produces exactly the same focused_element, visible_text, and url values as before. Downstream consumers (timeline aggregator, FTS indexer, content-dedup hasher) require zero changes.

[gemini-code-assist]

Code Review

This pull request introduces a modular app parser registry for the S1 enrichment pipeline, refactoring the existing logic into a more extensible architecture. It migrates browser-specific URL extraction to a dedicated parser and adds a comprehensive suite of unit, integration, and golden fixture tests. Feedback suggests implementing a deep merge for "app_context" to support better parser composition and switching to "read_bytes()" for JSON loading in tests to ensure robust encoding handling.