fix(daemon): poll every 25 ms on Windows - #33

Summary

• Part 5 of the stack for Idle daemon wakes the CPU ~100 times/s (10 ms sleep in the accept loop) #25, stacked on fix(daemon): wait for readiness instead of polling #31. Please merge fix(daemon): block subscriptions instead of polling #28 → fix(daemon): survive thread spawn failures #29 → refactor(daemon): route stop and ownership through Control #30 → fix(daemon): wait for readiness instead of polling #31 first.
• Interim, Windows-only reduction of idle wakeups. Unix behavior is unchanged.
  • The main loop's Listener::wait stub in src/transport/windows.rs now sleeps up to 25 ms instead of 10 ms (POLL_INTERVAL).
  • The per-terminal root-exit check in src/service/windows.rs now polls every 25 ms instead of 10 ms (EXIT_POLL_INTERVAL).
  • Together these take an idle Windows daemon from ~100 to ~40 wakeups/s in the main loop, and the same for each live terminal.
• Tradeoff: on Windows, a new connection or a terminal exit may now be noticed up to 25 ms later (it was 10 ms).
• The README, the transport/mod.rs doc, and the code comments now say 25 ms.
• Follow-up (the proper fix): have wait call WaitForMultipleObjects on the pending connect event plus a stop event, and have each terminal block in WaitForSingleObject on its process handle.
• Windows test timing: every Windows test deadline is 3 s or longer, and the src/daemon/server.rs timing tests are Unix-only, so no test needed changing.

Validation

• cargo fmt --check and cargo clippy --locked --all-targets --all-features -- -D warnings pass (toolchain 1.90.0).
• The macOS CI test selection passes: cargo test --locked --lib --test service --test runtime --test ownership --test rows --test ghostty-effects (lib 59, ghostty-effects 5, ownership 8, rows 1, runtime 8 + 1 ignored, service 9).
• Cross clippy with -D warnings passes for x86_64-pc-windows-msvc and x86_64-unknown-linux-gnu (--all-targets --all-features, ghostty-vt link libs stubbed). Native Windows and Linux are verified by CI.