Skip clipboard image capture for models without vision support

[yxlyx]

Problem

When the selected model does not support image input and a user pastes an image placeholder ([Image]), the UI still attempts to capture an image from the clipboard. If capture has not succeeded, it shows:

no image on the clipboard — copy an image first (this is Ctrl-V; ⌘V can't be captured)

This is misleading for non-vision models: it asks the user to go back and copy an image even though that model cannot accept one anyway.

Expected behavior

Check the selected model's image/vision capability before starting clipboard image capture. If the model cannot process images, immediately tell the user that image input is unsupported and do not show the clipboard-specific “copy an image first” message.

Acceptance criteria

• Pasting [Image] while using a non-vision model immediately shows an “images are not supported by this model” message.
• The clipboard capture flow is not started for non-vision models.
• The “no image on the clipboard — copy an image first” message is shown only when the selected model supports images but no image was captured.
• Vision-capable models retain the existing image paste/capture behavior.

[justrach]

Fixed on release/v0.0.217 (f580cc7), in PR #297.

Picks up @yxlyx's fix from codex/issues-258-260-integration, which could not land as written — it pushed src/readline.zig to 646 lines against the hard 600-line cap, so CI would have rejected it. That file was already at 599/600 on main, one line from the ceiling, which is why nothing could be added to it at all.

Rather than shave the patch, the file is split along its natural seam:

• The paste decision (ClipboardPasteSource, clipboardPasteSource, pasteMessage) moves to vision.zig, which already owns every dependency it has (visionCapable, StageResult, stageImagePath, grabClipboardImage). The three outcome strings collapse into pasteMessage(), and the "can't see images" sentence becomes one shared constant instead of two copies.
• HistoryNav + its two tests move to readline_history.zig — pure over std, so directly testable. readline.zig drops to 537 with real headroom.

Worth recording: a plain @import was not enough to keep the moved tests alive. The intermediate state read 351/351 — identical to main's baseline — because +2 new vision tests exactly masked -2 silently dead HistoryNav tests. Caught by baseline diff rather than by trusting the pass line; final count 353 on the branch, 363 on the release.

[justrach]

Fixed and released in v0.0.217 (commit f580cc7).

clipboardPasteSource checks vision support before touching the clipboard (src/vision.zig:152-156), so a non-vision model never triggers a capture. Both entry points go through it: Ctrl-V (src/readline.zig:262-274) and /paste (src/commands_model.zig:554-558).

Pinned by tests at src/vision.zig:248-278, which assert the clipboard grabber is never invoked on a non-vision model.