fix(tools): publish NID database cache after download completes

[darhebkf]

What

An interrupted download leaves a partial NID database that subsequent runs reuse. Download into a closed temporary file in the cache directory, publish it only after success, and clean up on failure. Register the four failure, retry and successful-cache regressions with CTest.

Tested

Rebased onto main d70b8998. macOS 15.7.3, Apple Silicon, AppleClang 17, Python 3.14.6. nid_database_cache passed through CTest, covering four regressions. The new group took 0.10 s locally.

The shared upstream C++ relinker sources were built in Release during this review; subsequent tool-only branches ran their affected CTest groups. Conventions and whitespace checks pass for each branch. Full Linux/Windows runtime validation requires a maintainer-dispatched Build run; no title or homebrew was executed.

Checklist

• Based on current main; no other open PR implements the same fix.
• One topic per PR; this remains the same fix with its regression tests.
• No source comments added; comment text in test fixtures is string data.
• Unimplemented paths throw; no runtime exports or silent stubs added.
• No notes, investigation files or images added to the repository.
• No new third-party code.
• Generic behaviour, not specific to one title.
• New test groups finish well below 30 s locally; CI-runner timings remain unverified.
• Depends on: none.
• AI-assisted: yes.

[adriadam10]

Checked: the temp-file + replace flow leaves no partial cache on failure, the four regressions are registered in ctest, and nid_database_cache fails without the nid_names.py change. Full build + ctest green after the rebase (555/555).
Rebased onto main (test-list adjacency in CMakeLists.txt with pr_overlap_exports); merging once CI is green.