feat(visa-sponsors): Canada provider (ESDC positive LMIA employers)

[type0labs-dev]

What

Adds a ca visa-sponsor provider backed by ESDC's positive LMIA employers list — employers issued a positive Labour Market Impact Assessment, i.e. cleared to hire temporary foreign workers.

• Latest quarter resolved via the CKAN API (package_show), never a hardcoded URL: newest English YYYYQn…Positive… resource wins. Note: quarters have shipped as XLSX since 2021 (the CSVs on the dataset stop at 2020Q3), so a minimal sheet extractor (shared strings + inline values, ~70 lines) was added on top of fflate — promoted from transitive to a direct orchestrator dependency (0 new packages in the lockfile).
• Columns mapped by header name (Province, Stream, Employer, Address, Approved Positions) so the provider survives the layout drift this dataset historically shows.
• route = program stream (High Wage / Low Wage / etc.), typeRating = "LMIA positive employer (2025Q4: N approved positions)", city parsed from the address, province in the county field.
• Catalog + docs table updated.

Evidence

• Live run: 9,737 employers parsed from the real 2025Q4 workbook in ~4.6s (sanity: Shopify Inc., Ottawa, Ontario, High Wage)
• 6 unit tests, including an in-memory XLSX built with fflate that mirrors the ESDC sheet shape (title rows before header, shared strings, entity decoding, empty-header error path)

Checks

biome ci . clean · shared + orchestrator typechecks · provider tests 6/6 · same 5 pre-existing client test files fail on clean main in my env (locale/timing), unrelated. Docs table conflicts trivially with #662/#663 — happy to rebase whichever lands last.

🤖 Generated with Claude Code

[DaKheera47]

small finding on this one, other than the merge conflict

The workbook has multiple rows per employer, but search keeps only the first one. This means the positions and streams shown can be incomplete.

Maybe aggregate rows by employer? sum approved positions, combine unique streams, and use “Multiple locations” when needed.