fix(cli): terminal detection for WezTerm/Ghostty/Warp/Konsole + Windo… - #1083

Good work here — several of these are genuine, well-reasoned bug fixes rather than cosmetic churn:

• generateKittyImageSequence in cli/src/utils/terminal-images.ts: the old code repeated the full control block (a=T,f=100,...) on every chunk and never emitted m=0 on the final chunk, which is a real spec violation that could leave transmissions unterminated on strict kitty implementations. The rewrite (only first chunk gets full control data, subsequent chunks carry m= only, last chunk gets m=0) is correct and matches the kitty graphics protocol doc you linked. The new terminal-images.test.ts chunk-boundary assertions back this up well.
• generateITerm2ImageSequence's size= param bug (base64 length vs decoded byte length) was a real, findable-in-review defect; the fix and the accompanying test (size=5 not size=8 for 'aGVsbG8=') are both correct.
• The WezTerm/Ghostty/Warp/Konsole detection additions are low-risk, env-var-only, and tested.
• The spawnPowerShell fallback to pwsh in clipboard-image.ts is a reasonable idea for minimal Windows installs, but it's untested — there's no test exercising the ENOENT-then-pwsh path, and the silent timeout bump in readClipboardFilePathWindows (1000ms → 5000ms) isn't mentioned in the PR description, so a reviewer can't tell if that's intentional or incidental.

My main critique is scope: this bundles three fairly independent changes (protocol chunking fix, terminal detection, Windows shell fallback) into one PR. Splitting the clipboard/pwsh piece out with its own test would make it easier to port and verify in isolation. The test/setup-scm-loader.ts stub is a sensible, clearly-commented mirror-only shim and shouldn't be ported as-is.

Overall the core protocol fixes are worth taking; the Windows fallback piece would benefit from a test before landing.