fix(telegram): search only on click, stop search floods and lost Telegram sources

[silentbil]

Telegram search is rate-limited per account. The app searched it automatically from every source list, firing about 26 requests per episode, so lookups kept timing out, sometimes with a toast over playback.

• New setting in Telegram settings, on by default and synced to the cloud: "Only search Telegram when clicking the Telegram source". The source list shows a "Click to search Telegram" row, results appear as they are found, and the player never searches.
• Lighter search: one request per phrase instead of two. The four phrasings that found every result in testing run first, two at a time. The rest only run if those find nothing.
• A timed-out search is no longer cached as "no results", which used to hide Telegram sources for hours.
• One shared search per episode. Starting playback stops running searches, and no Telegram toasts appear during playback.
• Fixed: the player's addons-only source lookup overwrote the saved source list, so Telegram sources disappeared when reopening an episode.
• The Telegram library now logs errors only.
• The source list shows "still checking N addons" while a source is still loading.
• Added Hebrew translations for the new texts.

[ProdigyV21]

Thanks for working on this! Searching only when requested and reducing Telegram traffic are useful improvements. It merges cleanly, the Play build compiles, and all 22 existing Telegram tests pass. I found three remaining issues to fix before merging:

1. Request timeouts can still be cached as no results. TelegramClient.sendRequest() returns null after its 20-second timeout, but TelegramRepository treats that as an empty response. A movie lookup can therefore finish before the outer 30-second deadline and cache an empty result for hours. Please propagate an incomplete outcome from individual requests and never cache it as a completed search.

2. Unfiltered search can miss playable files. The new request uses filter = null, reads only the first page, then discards non-video messages. Newer text/photo messages can fill that page while matching videos are further back. TDLib documents this as a paginated, newest-first search. Please use bounded pagination or a targeted video/document fallback, while keeping the rate-limit protection.

3. Manual results can still disappear when reopening Sources. searchTelegramNow() updates the resolver cache/UI, not the source-list cache. Reopening can return the older addon-only list before checking Telegram's cache. With Telegram as the only provider, the no-addons early return also discards cached Telegram results. Please merge those results on every cache-return path, or update/invalidate the relevant list caches after manual search.

Please add regression tests for a per-request timeout, non-video first-page results, and reopening after a successful manual search (including Telegram-only). The existing tests don't cover these new paths. I haven't tested this build on a physical TV or live Telegram account. Thanks!

[silentbil]

@ProdigyV21 thanks for the review! All three are fixed in 4bde881:

1. Per-request timeouts: a request Telegram doesn't answer in time now marks the whole lookup incomplete. Whatever was found is still shown, but it's never cached as a completed search.
2. Cut-off first page: when a phrasing's unfiltered page has more results behind it, the phrasing is repeated with the video and document filters. Pages with nothing behind them still cost a single request, so the rate-limit protection stays.
3. Reopening Sources: Telegram results found by clicking are now added on every path that returns a saved source list (fresh, stale, disk and no-addons), so they survive reopening, including with Telegram as the only provider.

The phrase loop and the result cache moved into Android-free classes (TelegramPhraseSearch / TelegramResultCache), with regression tests for a per-request timeout, a first page full of non-video posts, and reopening after a click-search (including Telegram-only).

Tested on an Android TV emulator with a live Telegram account: the click-search found all 10 matching files for Special Ops: Lioness S1E1 in about 2s, reopening Sources kept them without a new search, and nothing was cached from incomplete lookups.

[ProdigyV21]

Thanks @silentbil for addressing all three review points and adding the regression coverage. I checked the updated changes against current main and added one small follow-up: publish each answered page immediately, so a stalled sibling query or filtered follow-up cannot hide files already found when the overall deadline expires. Request counts and the two-query concurrency limit are unchanged. Both new regression tests reproduced the old behavior and now pass; all 36 focused Telegram tests pass, along with GitHub's full Sideload test build, web/backend tests and web type check. Merged with your original commits and authorship preserved. I have not repeated the live-account/emulator test or tested this on a physical TV.