fix(rendering): handle 0-byte audio segment files gracefully in Audio::new and SegmentRecordings (#2069)

[Samarth1306w]

Issue

Closes #2069

Summary

Fixes editor opening failure (Unable to Open Recording — segment 0 / mic / Failed to open audio: End of file) when recording bundles contain empty 0-byte audio segment files (audio-input.ogg size = 0).

Changes

1. Updated Audio::new in crates/rendering/src/project_recordings.rs to check std::fs::metadata(path).len() == 0 and return an explicit error (Audio file is 0 bytes (empty)).
2. Added graceful fallback handling in SegmentRecordings for mic audio loading errors (matching system_audio), logging a warning and treating as None so visual video and camera tracks remain fully readable and editable.

Acceptance criteria

• Checks metadata.len() == 0 before invoking ffmpeg::format::input
• Treats 0-byte mic audio segment files gracefully without crashing recording view
• Allows display and camera video playback even if mic audio segment is 0 bytes

/claim #2069

Greptile Summary

The PR makes recording metadata loading tolerate empty or unreadable microphone segments so projects can still open with their video tracks. It also adds per-IP rate limiting to public analytics and guest-checkout endpoints and introduces a unit test checking that declared rate-limit identifiers are referenced.

• Rejects zero-byte audio before invoking FFmpeg.
• Degrades microphone loading failures to an absent audio track for single- and multi-segment recordings.
• Adds rate limits to analytics tracking and guest checkout.
• Adds a source-level contract test for rate-limit identifiers.

Confidence Score: 5/5

The PR appears safe to merge, with no concrete changed-code defects identified.

The audio fallback preserves video accessibility for invalid microphone tracks, while the new public-route rate limits follow the existing fail-open per-IP implementation.

Important Files Changed

Filename	Overview
crates/rendering/src/project_recordings.rs	Detects zero-byte audio and gracefully omits microphone tracks when audio metadata loading fails; no concrete regression was established.
apps/web/app/api/analytics/track/route.ts	Adds fail-open, per-IP rate limiting before analytics payload parsing.
apps/web/app/api/settings/billing/guest-checkout/route.ts	Adds fail-open, per-IP rate limiting before creating guest Stripe checkout sessions.
apps/web/tests/unit/rate-limit-ids.test.ts	Adds a unit test scanning web TypeScript sources for references to every declared rate-limit identifier.

_{Reviews (1): Last reviewed commit: "fix(rendering): handle 0-byte audio segm..." | Re-trigger Greptile}

Context used:

• Knowledge Base — Export and Rendering Pipeline
• Knowledge Base — Web App (apps/web)

[superagent-security]

🚨 Contributor flagged. Click here for more info: Superagent Dashboard

[richiemcilroy]

Thanks for taking a shot at this. Closing it after investigating #2069 in depth, for a few reasons:

1. The investigation showed the recording-side bug described in Studio mode + camera on macOS: every audio-input.ogg is 0 bytes; editor fails with "Failed to open audio: End of file" #2069 cannot occur (details in the issue), so the "Closes Studio mode + camera on macOS: every audio-input.ogg is 0 bytes; editor fails with "Failed to open audio: End of file" #2069" framing does not hold. The genuine defect is only the editor-side open failure.
2. The mic fallback idea is right, but this diff moves the mic load above the system-audio load. ensure_start_time is stateful: the first track examined seeds the start-time expectation for the segment, so reordering can turn a currently recoverable mismatch on legacy bundles into a hard display / Missing start time failure. It also ships without tests or verification, and the 0-byte pre-check in Audio::new is redundant since FFmpeg already errors on empty files.
3. The rate-limiting changes to the analytics and guest-checkout routes are unrelated to this issue and touch a payment path, so they need their own PR and review. Feel free to resubmit those separately.

The editor-side fix landed via #2075, which preserves the existing load order and adds regression tests for the encoder path.