[ptescher]

Each platform's results name its own screenshots (closes #77)

expectation-run wrote a later platform's screenshots under platforms/<name>/screenshots/, while its manual looked in the docs' shared screenshots/ under the same names the default platform uses. The manual showed a placeholder, or the default platform's picture.

• A later platform's per-test screenshots go to screenshots/platforms/<name>/<test>/ in the shared docs directory, and its results name them so.
• New: RunOptions::screenshots_under, Acceptance::screenshots_under() and Acceptance::run_options(dir). expectation-run --docs uses them.
• A later platform's result that still names screenshots/<test>/… is shown as a placeholder, in its own manual and where the default manual borrows its results.
• The default platform and content-addressed stores are unchanged.
• The kettle demo's SwiftUI and React Native screenshots are moved to their folders (not re-captured), and their results files name them there.

Picture a start that couldn't be set up (closes #80)

• When the reset or an arrangement fails, the runner still asks for a screenshot and keeps it as the failed start's, in the docs directory or the store.
• The setup error and failed verdict are unchanged, later steps are skipped, and attachments are still kept.
• With screenshots off, the driver is not asked.
• A failed step's picture is captioned "What it looked like when this failed." in HTML and Markdown. This applies to failed actions too.

Tests

• crates/remote/tests/platform_screenshots.rs runs the real expectation-run binary for two platforms with the same story and different images, regenerates with expectation docs, and checks old-style names give a placeholder.
• crates/harness/tests/setup_failure.rs covers reset and arrangement failure with both destinations, a camera that also fails, and screenshots disabled.
• cargo test --workspace and cargo clippy --workspace --all-targets pass locally.

🤖 Generated with Claude Code