feat: keep ANSI by default and impl PTY mode

[gaojunran]

this is a stacked PR and should wait for #406 to be merged.

[greptile-apps]

Greptile Summary

This PR introduces a pty = true daemon configuration option that allocates a pseudo-terminal for daemons that rely on isatty() or produce colour output only when connected to a TTY. It also normalises ANSI handling throughout the stack: ANSI escape codes are preserved in log files, stripped before ready_output / on_output pattern matching, stripped in the web log viewer, and conditionally stripped in the logs CLI command depending on the output terminal's colour support.

Confidence Score: 5/5

Safe to merge once the stacked PR #406 lands; implementation is correct and previously flagged issues are all addressed.

No P0 or P1 findings. Previously flagged issues (saturating_pow backoff,
stripping in PTY mode, #[cfg(unix)] guards, stale port.bats assertions) are all resolved in this diff. The single P2 note is a silent behaviour change for on_output_hook scripts receiving ANSI-stripped text, which is consistent with the PR's intent but not explicitly documented.

src/supervisor/lifecycle.rs — the on_output_hook output change and the merged stdout/stderr channel are worth a second read, but both are implemented correctly.

Important Files Changed

Filename	Overview
src/supervisor/lifecycle.rs	Major refactor: stdout/stderr readers merged into a single channel; PTY master reader wired up in the same path; ANSI stripping added for pattern matching; backoff overflow fixed with saturating_pow; \r stripping for PTY ONLCR behaviour added. Silent behaviour change: on_output_hook hook command now receives ANSI-stripped output.
src/supervisor/pty.rs	New file: wraps libc::openpty into a safe PtyPair struct; correctly acknowledges missing FD_CLOEXEC in comments but does not set it; both master and slave OwnedFd ownership is handled correctly.
src/cli/logs.rs	strip_ansi parameter threaded through all output paths (pager, direct, live-stream); strip_ansi = raw
src/web/routes/logs.rs	Both the initial HTML render and the SSE stream now strip ANSI codes before html_escape; correct ordering (strip then escape) prevents ANSI sequences appearing as visible HTML entities.
tests/test_e2e_pty.rs	Four focused e2e tests covering ANSI preservation in logs (pipe mode and PTY mode), TTY detection with pty=true, and no-TTY default behaviour.
test/port.bats	Assertions updated to match new PortConfig serialisation (port = 3000 / expect = [8080] / bump = 10); previously flagged stale assertions are now corrected.
src/pitchfork_toml.rs	pty field threaded through PitchforkTomlDaemonRaw → PitchforkTomlDaemon → RunOptions; deprecation warning now fires in both the presence-only and coexistence-with-new-port cases.
src/supervisor/state.rs	pty field added to UpsertDaemonOpts; Default derived so builder can use ..Default::default(), removing large boilerplate struct literal.

_{Reviews (11): Last reviewed commit: "feat: keep ANSI and PTY mode" | Re-trigger Greptile}

[gaojunran]

@jdx ready for a review!

on_output hook payload is stripped and error-path fall-through

• it is on purpose. we don't need to capture a line of output with ANSI, so they are always stripped. And we don't want to expose hook error

tests/test_e2e_pty.rs — fixed sleep durations

• it's enough even for ci. I've tested it for multiple times

[jdx]

could we just make this default behavior?

[gaojunran]

you mean keep ANSI as default or strip by default? Now default is strip @jdx

[jdx]

I think we should keep ansi as default unless we've disabled colors, then we should strip when outputting the logs not when storing them

[gaojunran]

1. log.ansi has been removed. We always store ANSI
2. The output part (CLI, Web, TUI) decides whether to strip ANSI