Emit multiple video segments from a single tape

Is your feature request related to a problem? Please describe.

A tape often documents a sequence of steps (install → configure → run → verify) that I'd like to embed as separate short clips in different places (per-step README sections, tutorial pages, docs). Today VHS emits exactly one artifact per Output, so the only options are:

1. Copy the tape into N separate files — duplicating Set blocks, prompts, and any prerequisite state, and letting them diverge over time.
2. Post-process the single artifact with ffmpeg using hand-counted timestamps — which drift every time I edit the tape.

Both fight against the "one tape = one reproducible session" model that makes VHS pleasant to use.

Describe the solution you'd like

A new Split "<path>" command that closes the current video segment at the point it's executed and writes it to the given file. The extension picks the encoder path (.mp4 / .gif / .webm), matching how Output already works.

Output all.mp4              # optional: full recording still emitted

Type "brew install foo"
Enter
Sleep 2s
Split "install.mp4"

Type "foo configure"
Enter
Sleep 1s
Split "configure.gif"

Type "foo run"
Enter
Sleep 1s
Split "run.webm"

Semantics:

• Frames captured since the previous Split (or since recording started) form one segment.
• Segments can mix formats.
• Split and Output are independent — you can have neither, either, or both.
• Works cleanly with Hide/Show: Hide-period frames are not captured, so they don't affect segment boundaries.
• Boundaries are frame-exact and non-overlapping, so they don't drift on tape edits.

Describe alternatives you've considered

1. Multiple Output lines — would redefine "final file for the whole tape" to also mean "chapter delimiter", breaking existing tapes.
2. Post-process with ffmpeg -ss/-to — works but timestamps drift every time the tape is edited (add a Sleep and everything shifts).
3. Multiple tape files sharing state via Source — each tape still needs its own Set header and there's no clean way to preserve terminal state (cwd, env vars, running processes) across tape boundaries.

Additional context

I have a working prototype in a fork: tsuzu/vhs @ feat/split-command. Diff is ~170 lines across parser, executor, and render; all existing tests pass.