Badge composite frames on the film strip

[marcinz606]

Why

NegPy builds one frame from several files four ways — RGB triplet, stitch, HDR merge, and the reverse, the half-frame split. None of it shows on the contact sheet. The thumbnail grid paints the pixmap and nothing else, and the display name that carries (HDR) / (Stitch) / [1] is never drawn. So a merged frame looked exactly like a plain one, and the right-click menu was the only way to find out.

What

A badge in the bottom-left corner of the thumbnail, one glyph per kind:

Glyph	Frame
Two overlapping panes	stitched composite
Three stacked bars	merged bracket (HDR)
Three red/green/blue dots	RGB-scan triplet
Split box, one side filled	one half of a half-frame scan — the filled side says which

Plus a tooltip line with the count: HDR merge of 5 exposures, Stitched composite of 3 frames.

The chip is neutral grey, not the triage red. Red already means "you marked this" (bottom-right keeper/reject) and "this failed" (top-right decode error) in this delegate; what a frame is built from is a fact about the asset, not a state the user set. Each corner now owns exactly one meaning, and bottom-left stays clear of the keeper check at both ends of the thumbnail-size slider.

The one non-obvious bit

composite_kind() tests stitch before rgb. A stitch built from triplets also carries the primary part's green_path/blue_path — _on_stitch_registered copies them so thumbnail decode can find them — so the reverse order badges a stitch as a triplet. There is a test pinning this.

Changes

• negpy/desktop/session.py — composite_kind() / composite_summary() next to the other asset-dict readers; AssetListModel.data appends the summary to ToolTipRole. Failure tooltips untouched.
• negpy/desktop/view/sidebar/files.py — _draw_composite_badge() beside the two existing badge painters, same r=9 / inset-4 geometry. Called on the normal paint path and on the failed-decode placeholder card, so a composite whose part went missing still says what it is.
• docs/USER_GUIDE.md — a "Reading the badges" table under Triage covering all four corners and all four glyphs, plus a clause in the RGB Scan, Half Frame, Stitching and Merging sections.
• docs/PIPELINE.md — unchanged, no pixels move.

Tests

12 new cases in tests/test_file_browser_widget.py (nothing touched the delegate before): composite_kind per asset shape, the stitch-of-triplets precedence case, the summary strings, the tooltip through the real model, and paint checks that each kind marks the bottom-left corner and that no two kinds draw the same glyph.

Verification

• make lint, make type, make test: pass. One failure, test_stitch.py::test_stitch_real_rgb_triplet_samples — a PreviewManager.load_linear_preview_rgb() signature mismatch that fails identically on a clean main. Pre-existing, unrelated, not fixed here.
• Glyphs checked by rendering the real delegate offscreen over highlight, shadow and mid-grey thumbnails, at both ends of the thumbnail-size slider. Not yet exercised in a live make run session against real stitched/merged frames.