not an optimization engineer here, but I tried to collect the low hanging fruits

Measured locally on macos (:p), release build. Here is the comparison between baseline and this PR.

Idle (sleep 90s):

• CPU: 51% -> 6%
• RSS: 235 MB -> 208 MB

Burst (20k lines output):

• CPU: 47% -> 3%
• RSS: 237 MB -> 207 MB

Binary: 98 MB -> 76 MB

What changed:

• Switched bevy to explicit features instead of default-features = true
• Added WinitSettings with reactive low power mode
• Redraw throttle (16ms)
• Skip unnecessary work in Flat2d mode
• Reduce default scrollback (10000 -> 2000)
• Bounded PTY channel + larger read buffer

fyi, most of the diff is Cargo.lock (deleted unused bevy deps).