[Bug] Deleted PR files are marked complete without base content

[bitloi]

Description

PR file fetching stores metadata for deleted files, but skips content fetching for every file with status = "removed". Deleted files still have meaningful base-side content for scoring: the file existed before the PR and is removed at the PR head.

Because deletion-only PRs return successfully from the fetcher, the worker can set scoring_data_stored = true while the public files API returns the deleted file with base_content = null, head_content = null, and no byte size. Validators then see the PR as fully fetched even though the token input for the deleted file is missing.

Steps to Reproduce

1. Mirror or backfill a PR that deletes a file, for example a file with status = "removed" in the GitHub PR files API.
2. Let the fetch-pr-files worker complete for that PR.
3. Query GET /api/v1/pulls/:owner/:repo/:number/files.

Expected Behavior

Deleted files should get a pr_file_contents row with the base blob populated and the head blob left null.

For a removed file such as src/old.ts, the response should include:

{
  "filename": "src/old.ts",
  "status": "removed",
  "base_content": "<content from the merge-base/base side>",
  "head_content": null,
  "byte_size": 123
}

Actual Behavior

Removed files are filtered out before GraphQL content fetching:

const scored = files.filter((f) => f.status !== "removed");
if (scored.length === 0) return;

For deletion-only PRs, no content request is made, no pr_file_contents row is written, and the worker still marks file scoring data complete. The files API left-joins missing content and returns null content fields for the deleted file.

Environment

• OS: N/A
• Runtime/Node version: Node 20
• Browser (if applicable): N/A

Additional Context

Relevant paths:

• packages/das/src/webhook/github-fetcher.service.ts
  • fetchAndStorePrFiles stores removed file metadata in pr_files.
  • fetchAndStoreBatchedContents filters out removed files before content fetching.
  • fetchContentBatch already knows how to fetch base content for non-added files.
• packages/das/src/queue/fetch.processor.ts
  • marks scoring_data_stored = true after the fetcher returns.
• packages/das/src/api/pulls/pulls.service.ts
  • exposes file metadata and content through a left join to pr_file_contents.
• packages/db/09_pr_file_contents.sql
  • documents this table as PR file contents for token scoring: base and head versions.