fix: emit EventError once after all retries; skip retries for 4xx errors

[SuperCoolPencil]

Greptile Summary

This PR revises download failure handling and scheduler shutdown behavior.

• Emits the final error event from the scheduler only after retries are exhausted.
• Treats most HTTP 4xx responses as permanent failures while retaining retries for 429 responses.
• Preserves incomplete downloads and concurrent-download progress for later resumption.
• Moves potentially blocking progress sends outside the scheduler mutex and closes the shutdown signal before waiting for workers.
• Adds coverage for permanent HTTP errors and single error-event emission.

Confidence Score: 5/5

The PR appears safe to merge with no blocking failures remaining.

The scheduler-lock, shutdown, and stale-filename error-event failures are addressed without an eligible distinct regression.

T-Rex Logs

What T-Rex did

• Ran the targeted Go tests for internal/types and internal/scheduler with a focused run of Test(IsPermanentHTTP|Worker_SkipsRetriesOnPermanentError|RunDownload_DoesNotEmitEventErrorOnFailure), using -count=1 -v and a GOCACHE path.
• Verified the current run reports that internal/types passes HTTP status handling for permanent (400, 401, 403, 404) and retryable (429, 5xx) cases, and that internal/scheduler passes both lifecycle tests.
• Compared the current results with the pre-change runtime, which showed RunDownload emitting an EventError and four EventErrors for a permanent 403 failure, and observed that the current implementation completes with a single terminal scheduler-owned EventError.

_{Ran code and verified through T-Rex}

Important Files Changed

Filename	Overview
internal/scheduler/scheduler.go	Reworks retry classification, queue accounting, error-event emission, backoff, and shutdown ordering without leaving an eligible blocking failure.
internal/scheduler/manager.go	Removes per-attempt error emission and avoids discarding partial progress during concurrent-to-single fallback.
internal/orchestrator/events.go	Preserves incomplete download files when processing terminal error events.
internal/strategy/concurrent/downloader.go	Marks qualifying 4xx bootstrap responses as permanent HTTP failures.
internal/strategy/single/downloader.go	Marks qualifying 4xx download responses as permanent HTTP failures.
internal/types/errors.go	Introduces permanent HTTP error classification, excluding rate-limited responses.

Sequence Diagram

sequenceDiagram
participant W as Scheduler worker
participant D as Downloader
participant P as Progress channel
participant S as Graceful shutdown
W->>D: RunDownload
D-->>W: Return error
alt retryable and attempts remain
    W->>W: Requeue task as in-flight
    W->>P: Emit queued event outside mutex
    W->>W: Clear in-flight state and back off
else permanent or retries exhausted
    W->>P: Emit one error event outside mutex
    W->>W: Complete task accounting
end
S->>P: Close progressDone
S->>W: Wait for completion

Loading

_{Reviews (5): Last reviewed commit: "fix: keep tasks in-flight during retry p..." | Re-trigger Greptile}