feat(report): catalog KKSOLSW1, and make a skip fail where the feature exists

[BitHighlander]

R-4.1 — provider-attested Solana lookup-table accounts — had four tests and zero presence in the atlas. It's a headline 7.15 feature, and an auditor reading the PDF would have found no evidence it works. That's the exact failure the atlas exists to prevent.

S26–S29 catalog them:

S26	attested accounts are shown and the blind-sign warning survives
S27	a signature that does not verify changes nothing
S28	an attestation cannot be replayed onto another transaction
S29	with no signer loaded, a well-formed attestation is inert

S26 declares screens, so the filter picks it up and the screenshot leg captures it — the atlas is the single source of truth for OLED capture, so cataloguing it is what makes the evidence exist.

The Solana background now states the gap this closes. It described the 44-character address fix and stopped, which left S24 ("v0 with address-table lookups requires AdvancedMode") reading like a design choice rather than the open problem it is: the device cannot resolve a table it has never seen, so it routed those transactions to the blind-sign gate and signed accounts it never showed.

MUST_RUN_MODULES becomes version-aware. It was a flat set, so listing a 7.15-only module would have failed every older-firmware run for a feature that legitimately cannot exist yet. Now module → the version from which a skip becomes a failure. test_msg_solana_lut_attestation is listed at 7.15.0: all four gate on requires_message('LoadClearsignSigner'), so if provider loading regressed they'd all skip and the report would certify a feature it never exercised.