Add image and audio to prompt/exec

[vaibhavqrcg]

Adds support for multiple prompt content types in prompt and exec: text (existing), image, and audio. Users can pass files with --image and --audio (repeatable). Content is sent as ACP content blocks (text, image, audio) in session/prompt requests.

Usage

• Flags: --image <path>, --audio <path> (global; repeatable). Must appear before the command.
• Examples:
  • codeye --image screenshot.png exec "what is shown here?"
  • codeye --image a.png --image b.png cursor prompt <session-id> "compare these"
  • codeye --audio meeting.wav exec "transcribe and list action items"
• Order: Text block first (from positional args), then image blocks in order, then audio blocks. Image/audio files are read and sent as base64 with MIME type from extension.

Implementation

ACP

• PromptTextPart replaced by PromptPart with Type, Text, MimeType, Data (base64).
• TextPrompt(text) []PromptPart helper for text-only prompts.
• SessionPromptRequest.Prompt is now []PromptPart.

Client & runtime

• Client.Prompt(ctx, sessionID, parts []acp.PromptPart); all session runtime prompt/exec entrypoints take []acp.PromptPart instead of a single string.
• Existing call sites use acp.TextPrompt("...") for backward compatibility.

CLI

• internal/cli/prompt_parts.go: BuildPromptParts(text, imagePaths, audioPaths) builds the parts list, reads files, infers MIME type, and base64-encodes.
• Supported: images .png, .jpg, .jpeg, .gif, .webp; audio .wav, .mp3, .mpeg, .ogg, .flac, .m4a.
• dispatch for prompt and exec builds parts from positional text plus flags.AudioPaths and flags.ImagePaths.

Queue

• Request.PromptParts []acp.PromptPart (optional). When present, server uses it; otherwise uses Prompt as a single text part (backward compatible).
• Handler.Prompt(ctx, sessionID, parts []acp.PromptPart) updated; queue tests adjusted.

Docs

• README: --audio / --image in global options and a short usage note.
• SKILL.md: Same options plus an “Image and audio in prompt/exec” section with examples (one/many image, one/many audio, image+audio, for both exec and prompt).

Notes

• Agents must advertise image and/or audio prompt capabilities in initialization for those blocks to be accepted (per ACP).
• All tests updated and passing; existing behavior (text-only) unchanged.