Bug: Dashboard Chat — Ctrl+V paste broken, image paste not supported

Description

In the web dashboard Chat tab (launched via hermes dashboard --tui), clipboard paste has two issues:

1. Ctrl+V does not paste text — it is passed through to the TUI backend, which attempts to read from the server-side (WSL/Linux) clipboard. Since the browser clipboard and server clipboard are separate, this always results in "No image found in clipboard".

2. Image paste is not supported at all — even Ctrl+Shift+V (the documented paste shortcut) only calls navigator.clipboard.readText(), which ignores image data entirely.

Steps to Reproduce

1. Start the dashboard: hermes dashboard --host 0.0.0.0 --insecure --tui --no-open --skip-build
2. Open http://localhost:9119 in Firefox (or any browser)
3. Go to the Chat tab
4. Copy any text in the browser, press Ctrl+V in the chat input → "No image found in clipboard" error
5. Copy an image (e.g. screenshot), press Ctrl+Shift+V → nothing happens (image ignored)

Expected Behavior

• Ctrl+V should paste text from the browser clipboard into the chat input (standard browser behavior)
• Image paste (Ctrl+V or Ctrl+Shift+V with image in clipboard) should attach the image to the message, using the browser's navigator.clipboard.read() API to read ClipboardItem with image/png type

Root Cause

In web/src/pages/ChatPage.tsx (lines 345-386):

• attachCustomKeyEventHandler intercepts Ctrl+V and sends it to the PTY, instead of letting the browser handle it
• The paste handler (line 376) uses navigator.clipboard.readText() which only reads text — read() should be used to also handle image clipboard data

In tui_gateway/server.py (line 3410), the clipboard.paste RPC reads from the server clipboard, which is empty when accessed from a browser client.

Proposed Fix

1. Ctrl+V for text: Intercept Ctrl+V in the browser and use navigator.clipboard.readText() + term.paste() (same as current Ctrl+Shift+V behavior), so text paste works without round-tripping to the server
2. Image paste: When navigator.clipboard.read() returns image data (image/png, image/jpeg, etc.), convert to base64 and send to the backend via a new RPC method (e.g. clipboard.paste_image), or POST to an upload endpoint
3. Fallback: If the Clipboard API is unavailable (no user gesture, insecure context), fall through to the existing behavior

Environment

• OS: Windows 11 + WSL2 (Ubuntu)
• Browser: Firefox (latest)
• Hermes: latest (hermes-agent main branch)