fix(http): Last-Modified is the file's mtime, not now()

[vaibhav8a]

What

Last-Modified now carries the file's real modification time, and both filesystems agree on the comparison.

Fixes #1323.

Verified against your reproduction

last-modified: Tue, 31 Dec 2019 18:30:00 GMT     # the file's real mtime, not now()
last-modified: Tue, 31 Dec 2019 18:30:00 GMT     # …and stable 2s later

If-Modified-Since	status
exactly the mtime	304	✓
older than the mtime	200	✓
newer than the mtime	304	✓

How

FileSystem::modified_at reads the mtime through the same local-then-database fallback read_file uses, so the timestamp always describes the file actually served. DbFsQueries gains a prepared statement for the last_modified column, mirroring the existing was_modified / read_file / exists trio.

When no mtime is available the header is omitted rather than filled with a guess — your issue offers that option and I think it is the right one: a header saying "now" is worse than an absent header, because a client believes it.

On the comparison contract: >, and I got this wrong first

You said pick one and use it for both. I picked >= initially, reasoning it was the "safe" side.

Driving your reproduction proved that wrong. If-Modified-Since: T asks whether the file changed after T, so an mtime of exactly T is not a change and must answer 304. Under >= every revalidation re-sent the whole body — defeating the header this change exists to repair. The 304 row in the table above was a 200 until I fixed it.

So both sides now use >, which is the correct half rather than merely the chosen one. The DB query moves from >= to >; local stays >.

Scope

I did not add ETag or Cache-Control. Your issue lists them as optional, and they are a separate decision about caching policy rather than a correctness fix — happy to follow up if you want them.

Checks

• cargo build / cargo check — clean
• cargo clippy --all-targets — 0 warnings
• cargo fmt --check — clean
• cargo test --lib — 191 passed

(Local tests needed unixodbc installed to link; that is a machine setup detail, not a code change.)

[lovasoa]

Hi @ vaibhav8a ! The above looks like a raw llm output, feel free to edit it to explain your motivation in your own words, what you did and why, then switch the pr back to ready for review. You can explain why and how you used a coding agent, but I'm not accepting pull request that are not human reviewed.