[Tmwakalasya]

Why?

A Conduit response containing a requested diff as JSON null passes the map-key check but leaves a nil *diffResult. The provider's public Get method then panics when converting that diff to change information. This was reproduced with a local HTTP fixture containing one valid diff and one null diff.

What?

Treat a null requested diff like a missing diff at the response parser boundary. Return the existing missing-diff error before conversion, so a partially valid response does not panic or return partial change information.

Test Plan

• Added public-provider coverage for missing and null diffs alongside a valid diff. Before the fix, the null case panicked in convertToChangeInfo.
• ./tool/bazel test //submitqueue/extension/changeprovider/phabricator:go_default_test --test_output=errors passed, including the existing valid-response controls.
• make fmt, make lint, make check-tidy, and make check-gazelle passed.

This is defensive handling of a constructed response, not a reported production incident. The full integration/e2e suites were not run.

Issue

No linked issue.

[Tmwakalasya]

@behinddwalls Just investigated the CI failure , the React 18 runtime test hit its 5-second timeout. The Phabricator provider tests, build, and integration checks all passed, so this looks unrelated to my change. Could you please rerun the failed Unit Test job when you have a chance and thank you.