feat: make slash completion prefix-first with advisory arguments

[Brian Krabach (bkrabach)]

Summary

Make interactive slash completion prefix-first while keeping argument suggestions optional.

• Open unselected menus for known arguments as well as slash-command names.
• Make unselected Tab accept one unique match or extend only the longest common prefix. Repeated Tab does not cycle, and an exact short name remains ambiguous when a longer name also matches.
• Keep explicit arrow/Shift-Tab selection separate from acceptance. Tab, Enter, or Space accepts a selected item without executing the command.
• Prefer an exact typed command on unselected command-menu Enter, then a unique match; otherwise keep the menu open with a visible “Type more or choose” hint. Unselected argument-menu Enter submits the typed line unchanged, including with automatic popups disabled.
• Preserve literal Space editing, avoid duplicate completion delimiters, retain later text during midline completion, and render the ambiguity hint without requiring terminal cursor-position-request support.

The existing ui.slash_popup.enabled setting still defaults to true. Setting it to false disables automatic command and argument menus, not manual Tab completion or advisory argument submission. Command validation, completion metadata, and snapshot-only keypress callbacks are unchanged.

Verification

• uv run --with truststore pytest -q — 2,183 passed, 1 skipped, 13 deselected, 1 xfailed.
• uv run --with truststore pytest -q -m integration — 13 passed.
• uv run --with truststore pytest -q tests/test_interactive_slash_completion.py tests/test_interactive_chat_tty_teardown.py — 74 passed.
• Ruff checks on the changed Python files passed.
• Independent source and pre-publication reviews passed.
• Live DTU keyboard checks passed at 120×40 and 80×24: common-prefix completion, exact short-name acceptance, visible ambiguity guidance in CPR-unsupported terminals, separately paced selection/acceptance, advisory argument submission, popup-disabled manual completion, and clean exits.

The corrected live report retains an overall PARTIAL label because two text-readiness probes timed out. Fresh screenshots immediately afterward confirmed the live session title and prompt before the tested flows; all four corrected keyboard cases passed. The initial failure report is preserved separately. Exact trailing-space counts are covered by real PromptSession pipe tests rather than whitespace-trimming screen captures.

Scope and limitations

• Updates the implementation, regression tests, and interactive documentation only.
• No new settings schema, in-app UI controls, shell-completion changes, provider behavior changes, or new skill metadata.
• Backspace still closes the menu until another character or Tab; that existing limitation is unchanged.
• Private DTU artifacts, logs, screenshots, fixtures, and local verification directories are not included.