Feature request: support slash commands mid-prompt

Problem

Slash commands are only recognized at the very start of the input, which creates two daily friction points:

1. Skills and plugin commands can't be embedded in a natural prompt. If I want to give the skill context before invoking it — e.g. review this diff /skill:code-style with care — I can't: the command only activates when the message starts with it, so the text I already typed either has to be discarded or becomes a plain message with no activation at all.

2. Mode toggles and panels require a cleared input. Halfway through writing a long prompt I often realize I want Plan mode, a different model, or a permission change. Today that means clearing (or submitting) my draft, running /plan / /model, and retyping the prompt.

Proposal

Allow slash commands mid-prompt:

• Typing / after existing text offers completion for skill/plugin commands and a curated set of quick built-ins (toggles, pickers, info panels); other built-ins stay leading-only.
• Skill/plugin tokens in free text activate on submit, with the surrounding text as the command's input (exact-name match only, so /tmp stays plain text).
• Quick built-ins (/plan, /model, /yolo, …) run immediately when selected, leaving the rest of the draft in place.

A working implementation with tests and docs is available in #2496 (closed per the "discuss first" policy — happy to reopen or rework it based on your feedback).

Start by reviewing the working implementation, tests, and docs referenced in issue #2496, then compare them with this proposal's mid-prompt skill/plugin and quick built-in behavior. Done means the proposed command completion, activation, draft preservation, and leading-only exceptions are covered consistently by the implementation, tests, and documentation.