prompt.submit hook pipeline reworked with rewriting, an 'enter' phase, and better diagnostics

What

• Hooks that run on prompt.submit can now rewrite the text of the prompt before it is committed; this is logged as "prompt.submit: text rewritten by a hook".
• The pipeline gained a new enter stage/callback. When it fires, Claude Code now records telemetry event prompt_queued and the prompt is queued rather than run immediately.
• When a hook drops a prompt, Claude Code now logs a reason ("prompt.submit: dropped") and can show a distinct "Prompt not submitted" message alongside the existing "Prompt dropped by a hook" message.
• If a plugin's prompt.submit hook attaches extra context to its result after it has already called next(), that is now detected and a warning is logged that the entries were dropped, instead of the entries being silently lost.
• Internally, both the interactive and SDK/headless prompt-processing paths now open a disposable tracker before processing input and mark it "handed off" once the new messages are pushed into the conversation; if the tracker is disposed before hand-off, any waiter is rejected instead of hanging.
• The SDK's error message for $.prompt.submit (thrown when text starts with /) now tells the caller to use $.command.run instead.

Why These changes make the prompt-submission path more capable (hooks can now edit prompts, not just allow or block them) and easier to debug, since drops, rewrites, and queued prompts are now logged and reported distinctly instead of failing silently or hanging.