task: document the new sensor taskings, and fix two broken examples

[maximelb]

Summary

Documentation only — no functional change, and none is needed.

The CLI does not parse or validate the task string. task send forwards it to the backend verbatim (commands/task.py:127 → sdk/sensor.py:118), and the backend generates its own help from the command parser's argument definitions. So dir_find, file_grep, container_list and artifact_get --root-dir already work against the current CLI:

limacharlie task send --sid <SID> --task 'dir_find "C:\Users" -x "*.exe" --with-hashes'

What was missing is discoverability. This adds them to the task send command menu, which also feeds --ai-help through the explain registry.

I also noted the budget convention once: a bounded search stops at the first budget reached and reports it in SCAN_IS_TRUNCATED / SCAN_STOPPED_REASON, so an empty result is not necessarily a clean host.

Rather than inline ~15 flags per command — this menu has never carried flags for any command — the text points at the backend reference, which is the authority on them.

Two pre-existing bugs, same lines

Both verified against the actual Click options rather than assumed:

• The incident-response cheatsheet (help_topics.py) told users to pass --args to task send, which accepts only --sid, --task and --investigation-id. Both examples errored out on copy-paste. Rewritten as task strings, matching the form the reference documentation uses.
• doc/cli/sensor-management.md used task send --command. send takes --task; --command belongs to request and reliable-send.

Context

repo_list and usb_list_devices are also absent from this menu — this list has drifted from the backend for a while. I did not add them here to keep the change scoped, but they are worth a follow-up.

Validation

• pytest — 4005 passed, 5 skipped
• Confirmed the rendered --ai-help text and cheatsheet output, not just the source

[maximelb]

/lc-review