Give the backend calls a deadline

[ramonski]

http::get and http::post connected with TcpStream::connect and read with read_to_string, neither of which has a timeout. A backend that accepts the connection and then never answers blocks the caller forever.

The caller that matters is the tray ticker. self_heal_from_backend() runs inside spawn_ticker, which is tauri::async_runtime::spawn, so a hung read pins a Tokio worker and the menu-bar pill stops updating until the app is restarted.

That is the exact outcome the surrounding code already works to prevent. Two comments in spawn_ticker explain the catch_unwind guards:

a panic in the renderer […] just skips this tick instead of killing the whole ticker task. Without this guard a single bad frame freezes the menu-bar pill indefinitely until the user restarts the app.

The guard covers panics. A hang arrived through the other door, and a blocking socket read with no deadline is the more likely of the two for a local HTTP call.

toggle_timer has the same call on its own std::thread::spawn, so there it leaks a thread rather than freezing the tray.

Five seconds on connect, read and write. Generous for a local FastAPI answering /api/clocks/active, short enough that a stuck one costs a single tick.

Tests. The crate had none; it has two now. connect_to takes the address so the test drives the real function rather than a copy of it — with the timeouts removed, a_silent_backend_does_not_block_forever fails after 30 seconds instead of passing in 5.

test http::tests::a_silent_backend_does_not_block_forever ... ok
test http::tests::a_normal_response_is_parsed ... ok
test result: ok. 2 passed