AlmaCareer monitor + 18 pilots (supersedes #2443) (#2540)

* Add AlmaCareer monitor for jobs.cz (CZ) and topjobs.sk (SK)

Alma Career runs the Capybara-framework career sites used by many
employers in Czechia and Slovakia.  Both countries share one GraphQL
backend at api.capybara.lmc.cz — only the host, widgetId, and
per-tenant apiKey differ, so they are handled by a single `almacareer`
monitor type with a `country` metadata field.

The monitor:
- Extracts widgetId + apiKey from the tenant's /assets/js/script.min.js
  bundle (they are baked in at deploy time).
- Walks `jobAdList` pagination (server-capped at 10 items/page).
- Fires per-job `jobAd(id)` queries concurrently to pull full
  content.htmlContent (the list payload only includes a teaser).
- Registers as a rich monitor (scraper auto-skipped).

Closes #2434.

* Bump apps/crawler/VERSION to 0.8.65

Covers the AlmaCareer monitor addition.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Apply Stage 2 review polish + add 10 pilot companies

Polish (apps/crawler/src/core/monitors/almacareer.py):
- Harden _WIDGET_MAIN_RE: switch from balanced-brace body match to
  anchor-on-"widgets":{"main":{ + 4 KB bounded window; tolerant to nested
  objects ("themes", "filters") between "id" and "apiKey" in the minified
  bundle.
- GraphQL partial-data handling: prefer payload.data when both errors and
  data are present (Capybara routinely returns both for optional-field
  resolver failures); log "graphql_partial_data" and only raise if data
  is missing.
- Log _fetch_widget_config exceptions and non-200 responses under
  "widget_config_fetch_failed" instead of swallowing silently.
- Loosen _parse_location: keep up to five distinct parts (cityPart, city,
  district, region, country) instead of silently dropping district+region
  past two leading parts.
- gather(return_exceptions=True) on the detail fan-out so one flaky
  detail fetch doesn't poison the whole discover run; log
  "hydrate_failed" per-job.

Tests (apps/crawler/tests/test_almacareer.py, +6 cases):
- Empty listing returns [] without firing detail fetches.
- _fetch_job_html returning None preserves the listing teaser.
- can_handle page-HTML marker fallback (custom-domain portal finds tenant
  via cdn.capybara.lmc.cz + data-host).
- _match_country uppercase host.
- _parse_location with five distinct parts.
- _extract_widget_config with nested objects before "apiKey".

Pilot companies (10 boards via the new monitor):
- New CZ companies: albert-cz, csob, kaufland-cz, kb-cz, o2-cz.
- Additional boards on existing companies: bosch, ey, kpmg, siemens,
  vodafone (each gets a *-cz-almacareer board).
All 10 pilots verified with direct discover() probe (alma GraphQL only):
albert 173, kaufland 139, csob 107, o2 94, vodafone 90, kb 71, ey 45,
siemens 43, bosch 31, kpmg 23 — total 816 jobs.

apps/crawler/VERSION: 0.8.65 → 0.8.70.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Fill full metadata + R2-staged logos for 5 new CZ companies

* Upload images to R2

* AlmaCareer: hardening + 8 more pilots (5 CZ + 2 existing + 1 SK)

Review findings from a close read of the monitor + `can_handle` path,
plus eight additional live-verified pilots including the first
`topjobs.sk` tenant.

Code fixes
----------

- **``BoardGoneError`` on 404 `script.min.js`.** `_fetch_widget_config`
  previously swallowed every non-200 to `None`, which raised a generic
  `ValueError` in `discover` — the board would retry 5 times before
  auto-disabling. Now 404 → `_WidgetConfigGone` → `BoardGoneError`
  (one-shot). Transport errors propagate as retriable.
- **`_WIDGET_WINDOW_BYTES`: 4KB → 64KB.** The bundle is <300KB; the
  old window had thin margin if Alma ever enlarges `themes` / `filters`
  between the `main` anchor and `apiKey`. Single-pass regex; linear
  in window size.
- **`null` `jobAdList` raises.** GraphQL schema drift would otherwise
  surface as a silent 0-job crawl that eventually auto-disables the
  board. Now it raises so the failure is loud.
- **`o2-cz` logo_type: `lettermark` → `wordmark+icon`.** `lettermark`
  wasn't in the validator's enum, so the PR shipped with a dormant
  validation error that `inspect.validate_csvs()` now surfaces.

Pilots (live-verified from my IP 2026-04-23)
--------------------------------------------

Existing companies (new boards):
- `mcdonalds-cz-almacareer`   mcdonalds.jobs.cz       169 jobs
- `mcdonalds-sk-almacareer`   mcdonalds.topjobs.sk     18 jobs (first SK tenant)
- `bmw-cz-almacareer`         bmw.jobs.cz              26 jobs

New companies (wikidataId + SVG logo + icon + en description):
- `penny-cz`     Penny Market        273 jobs
- `kfc`          Kentucky Fried Chicken 180 jobs
- `billa-cz`     Billa               155 jobs
- `t-mobile-cz`  T-Mobile Czech       61 jobs
- `ikea-cz`      IKEA CZ               9 jobs

Rewe-group sibling retailers (Penny, Billa) carry a shared
``parentOrganization`` note; KFC is Yum! Brands; T-Mobile CZ is
Deutsche Telekom.

All logos sourced from Wikimedia Commons (SVG) + apple-touch-icons /
favicons from each tenant's home page.

Addresses critic's N2 (no SK pilot verified live) by including
`mcdonalds-sk-almacareer`.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Author: 3 people

VERSION

@@ -1 +1 @@
-0.7.4
+0.7.5

apps/crawler/VERSION

@@ -1 +1 @@
-0.8.75
+0.8.76