Query-driven search: resolve t=tvsearch against the broadcaster on demand (DR-011)

[ChrisonSimtian]

Description

t=tvsearch / t=search currently only finds what a crawler has already written to Postgres, and the
crawl list is three hardcoded seed shows (extra 3, Die Biene Maja, heute-show in
Agents/{Ard,Zdf}/Program.cs). So Sonarr searching for anything else gets an empty feed — indistinguishable
from "not available in the Mediathek".

That is backwards for a Newznab indexer. Per
DR-011, search should resolve against
the broadcaster on demand, which is how MediathekArr works: it is "pretending to be a usenet indexer, but
actually just fetching and parsing search results from MediathekViewWeb". Sonarr asks, the indexer goes and
looks.

This is the piece that makes Krautwatch usable with zero *arr configuration — no instance, no API key,
no pre-registered show list.

Scope

• On a t=search/t=tvsearch miss, resolve live via IBroadcasterCrawler for the queried title,
  persist what comes back, and serve it in the same response.
• Cache negative results too, with a shorter TTL. Sonarr retries the same failing query on a
  schedule; without this, every miss becomes a fresh multi-hop crawl of ARD.
• Bound the request. The ARD path is A-Z widget → show page → episode list → item page, which is far
  slower than Sonarr's expectations for an indexer. Needs a hard timeout and a partial-results answer
  rather than hanging the search.
• Be polite to ARD/ZDF: coalesce concurrent identical queries, and rate-limit outbound so a Sonarr
  library refresh cannot turn into a crawl storm.
• Keep the standing crawl list for the RSS feed (DR-011) — this issue changes search, not RSS.
• Decide whether a live-resolve miss should also add the show to the standing list, so RSS-Sync picks
  up future episodes automatically.

Usage Example

# Today: empty unless "Tatort" happens to be one of the seeded shows
GET /api?t=tvsearch&q=Tatort&apikey=…      → <rss> with no items

# Target: resolves against ARD on demand, caches, and returns
GET /api?t=tvsearch&q=Tatort&apikey=…      → <rss> with Tatort episodes

Alternative

Keep requiring every show to be pre-registered — either by hand in Crawl:Targets (RundfunkArr's model)
or pre-warmed from Sonarr (#6). Both work, but both mean an unregistered show silently returns nothing,
and neither behaves the way an operator expects an indexer to behave.

Notes

• Latency is the central design risk, and it is the reason to do caching and timeouts as part of this
  issue rather than after it. Putting broadcaster round-trips in Sonarr's request path is the trade DR-011
  accepts consciously.
• Matching quality decides whether results are usable at all. EpisodeNumbering.Parse and SeriesType
  exist; Show.TvdbId is reserved but never set. Both comparables invest heavily here — MediathekArr calls
  out "the horrendous lack of consistency and metadata in ARD/ZDF Mediatheken".
• Neither comparable documents how it serves RSS-Sync, so that half cannot be copied and still needs a
  standing crawl.
• Relates to Remove the dead MediathekView catalog subsystem (drops the SharpCompress NU1902 vulnerability) #49 (on hold): if per-show crawling proves too narrow, the MediathekView filmliste is the
  full-catalog alternative — which is exactly what MediathekArr leans on.