Mark HTTP/2 flows imported from HAR files as HTTP/2 - #8583

Description

The HAR importer stores HTTP/2 as "HTTP/2", but Message.is_http2 and the HTTP/2 layer use "HTTP/2.0". So HTTP/2 flows loaded from a browser HAR (Chrome writes http/2.0, Safari writes HTTP/2) have is_http2 == False. A HAR exported by mitmproxy itself writes "HTTP/2.0", which matches none of the cases and comes back as HTTP/1.1.

It shows up in client replay. With is_http2 False, the HTTP/1 layer skips its HTTP/2 to HTTP/1 conversion and puts the version straight into the request line. With a HAR entry that has "httpVersion": "http/2.0" and a plain python -m http.server as the target:

$ mitmdump -C h2.har
[replay]: GET http://127.0.0.1:18779/hello HTTP/2
 << HTTP/1.0 400 Bad request version ('HTTP/2') 487b

The server receives GET /hello HTTP/2. With this change the request goes out as GET /hello HTTP/1.1 and gets a 200.

This maps http/2.0, HTTP/2 and HTTP/2.0 to HTTP/2.0 for both the request and the response. The expected http_version in five HAR fixtures changes to match (head-content-length.har was exported with HTTP/2.0, so it used to load as HTTP/1.1).

Chrome HARs also list the HTTP/2 pseudo-headers (:authority, :method, ...) as ordinary headers, and the importer keeps them, so replaying a Chrome entry still fails, on main and on this branch alike, with Received invalid header name: b':authority'. I left that for a separate PR, along with Chrome's h3 version string, which still imports as HTTP/1.1. I can send either.

#8535 edits the same match blocks for HTTP/1.0. The two changes don't depend on each other, and whichever lands second needs a one-line rebase.

Testing:

• New test_http2_version checks is_http2 on both sides for all three spellings. It fails on main with assert False where False = Request(GET address:22/path).is_http2.
• The replay above, before and after.
• tox -e py,lint,mypy,filename_matching passes on main and on this branch (2026 vs 2029 passed, the 3 new tests), and individual_coverage is ok for mitmproxy/io/har.py.

Checklist

• I have updated tests where applicable.
• I have added an entry to the CHANGELOG.