[SonAndE14]

Add explicit multi-source and All compatible source selection while preserving the original single-source picker. Preserve source identities and row ownership for pagination. Show partial results, bound catalogue concurrency to three and serialize calls to the same provider, retaining successful rows when another provider fails or times out. Publish rows before waiting for featured metadata; bound featured and recommendation work separately.

Offer six or twelve recommendations with All/per-source filtering. Start with native related titles; consult matching titles in selected direct providers and then fill from same-type catalogue suggestions if necessary. Catalogue suggestions are not claimed to be semantically related recommendations. Exclude the current title and presentation duplicates without merging histories. Do not fabricate twelve results when the sources do not supply enough.

Related to #300. Regression tests cover partial results, ordering, cancellation, concurrency and recommendation selection. Physical pagination, D-pad focus, source behavior and latency measurement remain pending. Card scaling and featured-count presentation are submitted separately.

Validation: :library:checkKotlinAbi, :app:testPrereleaseDebugUnitTest, and :app:assemblePrereleaseDebug passed locally. 15 JVM tests passed without failures or errors. Lint is left to this PR's GitHub workflow; the complete integration branch previously passed Lint. These results are not physical-device validation.

AI disclosure: implementation, code review and automated test execution used OpenAI Codex assistance, as required by AI-POLICY.md. This draft submits the implemented work for human review; pending physical checks and limitations above are not claimed complete. No changes to provider source code or the projector launcher are included.

[fire-light42]

The code is not up to standard, and way too bloated.