fix: stop swallowing exceptions in import fire-and-forget pipeline runners

[davidortinau]

Problem

Three fire-and-forget Task.Run blocks in the import/channel endpoints had empty catch bodies with a misleading comment:

catch (Exception ex)
{
    // Logging happens inside the pipeline service
}

This comment is incorrect for exceptions thrown before the pipeline service's own try/catch (e.g. DI resolution failures, scope creation errors, the first pipeline step). When that happens:

• ex is silently discarded
• The import record is never updated to Failed
• The caller polling /api/imports/{id} sees the import permanently stuck in Pending

Fix

Replace the empty catch with logging + status update in all three call sites:

catch (Exception ex)
{
    logger.LogError(ex, "Unhandled exception in import pipeline for import {ImportId}", import.Id);
    await pipelineService.FailImportAsync(import, ex.Message);
}

Changes

• VideoImportPipelineService.cs — promoted FailImportAsync from private to public so endpoint handlers can call it
• ImportEndpoints.cs — StartImport and RetryImport catch blocks now log the error and mark the import failed
• ChannelEndpoints.cs — channel video trigger loop catch block gets the same fix

Verification

• dotnet build src/SentenceStudio.Api/ — ✅ clean (0 errors)
• dotnet test tests/SentenceStudio.Api.Tests/ — 120 pass; 19 failures are pre-existing on main (IChatClient not registered in auth test DI harness, unrelated to this change)