simulate: list, view and export subcommands

[u9g]

Requested in Slack: lk agent simulate --view can reopen a past run, but there was no way to list runs to find an ID.

Rather than add --list next to --view and --export, this makes the run operations subcommands. Each of those flags took an ID and replaced the whole run action, and audio inherited them, so lk agent simulate audio --view ID parsed and did something meaningless. A flag that swaps the action is a subcommand; this matches the noun/verb shape of the rest of lk (lk agent list, lk room list, lk egress list) and the git stash / git stash list / git stash show precedent of a verb whose bare form does the main thing.

lk agent simulate                     prints the subcommand list, runs nothing (was: text run)
lk agent simulate text [entrypoint]   run, text mode (new; was bare simulate)
lk agent simulate audio [entrypoint]  run, audio pipeline (unchanged)
lk agent simulate list                newest runs for the project (new)
lk agent simulate view <run-id>       was --view
lk agent simulate export <run-id>     was --export

Commit 1 turns --view/--export into view/export with a positional run ID and updates the exit hints to print the new forms. The old flags are removed, not aliased: the hint is reprinted after every run, so only stale scrollback breaks.

Commit 2 adds list: one ListSimulationRuns call for the project's first page (server returns newest first, 20 per page, no jobs), rendered like lk agent list with --json for the raw response.

Commit 3 adds --limit to list, the same next_page_token loop as lk egress list: unset keeps the single-page default, --limit N walks pages until N runs are collected. --json prints the collected runs inside the List response so the field formatting is unchanged. No filters yet.

Commit 4 adds text: simulate audio names its mode, so simulate text should exist too.

Commit 5 makes bare simulate a help command. With list/view/export under it, a verb that also runs on its own is the odd one out, and lk agent simulate typed alone now shows the options instead of starting a run. The project is not resolved on the help path. Docs that say lk agent simulate for a text run need to say lk agent simulate text.

The run flags on simulate stay persistent, so list/view/export still parse --scenarios etc. and ignore them. Making them local would break lk agent simulate --scenarios f audio: urfave/cli v3 resolves a flag by the first command in the lineage that declares it and re-runs PreParse on the child's flags, so a shared or duplicated flag loses the value parsed on the parent.

Tested against my dev project: list, list --limit 3, list --json, view <id> (CI and TUI paths), export <id>. Paging tested against a staging project with 184 runs: --limit 45 returned 45 unique runs across three pages, newest first with no seam; --limit 1000 drained all 184.

[u9g]

Demo of the subcommands against my dev project: bare lk agent simulate printing help and running nothing, text --help, list, list --limit 3, list --json, export, view in CI mode and in the TUI, and audio --view failing to parse.

simulate-subcommands.mp4

[theomonnom]

lgtm!