cppa-cursor-browser: Cache invalid-workspace alias inference on workspace context

[clean6378-max-it]

Calendar Day

Week 5 carryover from PR #113 (PR 1 of 2 — deferred review item 1c)

Planned Effort

5 story points (Medium–High) — sprint item #1c

Problem

When invalid_workspace_ids is non-empty, assemble_single_tab runs a full global scan (COMPOSER_ROWS_WITH_HEADERS_SQL) on every deep-link request solely to build invalid_workspace_aliases via infer_invalid_workspace_aliases. List, summary, and full /tabs paths already pay this cost once per request; single-tab loads repeat it for every GET /api/workspaces/<id>/tabs/<composer_id> call. On large Cursor installs this makes search deep-links and conversation URLs noticeably slow.

The same alias map is recomputed independently in list_workspace_tab_summaries, assemble_workspace_tabs, _load_search_workspace_assigner, workspace_listing, and export_engine whenever invalid workspaces exist.

Goal

Compute invalid_workspace_aliases once per storage fingerprint (same mtime scheme as summary_cache / composer-map cache) and expose it on WorkspaceContext (or a sibling cached field) so hot paths — especially assemble_single_tab — read a precomputed map instead of scanning all composerData:* rows.

Scope

Touch points

• services/workspace_context.py — extend WorkspaceContext (or enrich helper) with invalid_workspace_aliases: dict[str, str]
• services/summary_cache.py — fingerprint include/exclude rules consistent with existing composer-map cache
• services/workspace_tabs.py — assemble_single_tab: consume cached aliases; remove per-request COMPOSER_ROWS_WITH_HEADERS_SQL roster scan
• Call sites that currently call infer_invalid_workspace_aliases inline when invalid_workspace_ids — switch to context cache where fingerprint matches
• Tests: tests/test_workspace_tabs_summary.py (scoped load assertions), new test that single-tab does not run full composer SQL when aliases are cached

Out of scope

• Loading full global bubble_map on list/summary (separate perf tradeoff, Project list and workspace details show different conversation counts; workspace load still slow on large installs #95)
• Changing alias vote/tie-break semantics in infer_invalid_workspace_aliases

Acceptance Criteria

• assemble_single_tab no longer runs COMPOSER_ROWS_WITH_HEADERS_SQL on every request when invalid workspaces exist and cache is warm
• Alias map invalidates when workspace storage fingerprint changes (same as other summary-cache entries)
• Existing alias-resolution behavior unchanged (same inputs → same invalid_id → replacement_id map)
• test_scoped_mrc_load_with_invalid_workspaces and related tab tests still pass
• New or updated test proves single-tab path avoids full composer scan when cache hit
• mypy --strict and full pytest pass
• PR approved by at least 1 reviewer

Verification

cd C:\Users\Jasen\CppAliance\cppa-cursor-browser
.\.venv\Scripts\Activate.ps1
pytest tests/test_workspace_tabs_summary.py tests/test_workspace_list_count_alignment.py -q
pytest -q
mypy .