ISO 9660: non-zero-origin coverage completes ADR-0005 per-backend

[bmxcode]

What this finishes

ADR-0005 — "probe for the filesystem origin; never assume byte 0." The failure it exists to prevent is silent: a parser assuming byte 0 reads a zeroed pregap, finds no volume, and reports an empty disc, not an error.

#49 added synthetic non-zero-origin tests for EMU3 and Roland (and, in doing so, uncovered and fixed a real EMU3 bug where scans returned file-relative addresses, so at a non-zero origin every bank listed and read back empty). ISO 9660 was the one filesystem backend left without such a test — and it is the exact hybrid-disc case the ADR names: an ISO track ahead of the sampler partition, or a pregap left inside the cooked stream.

What the test now proves

test_origin_resolves_when_the_pregap_is_inside_the_cooked_stream prepends a 150-sector zeroed pregap to a synthetic make_iso9660 image, resolves with find_origin, and asserts:

• the probe resolves origin.offset == 150 * 2048 and backend.name == "iso9660" (not an assumed byte 0);
• the volume label and file listing resolve from that origin;
• read_file returns the payload byte for byte from the resolved origin.

The read-back is the assertion that matters. A listing alone passes even when the reads are wrong — an extent-vs-origin bug can list every file and still return pregap zeros, which is exactly how the EMU3 bug hid behind a whole-looking listing. This is the discriminating property the pre-#49 tests lacked.

Was the ISO 9660 path offset-clean? Yes — no fix needed

Every read the backend does is origin + extent * SECTOR_SIZE: probe, the descriptor scan, the directory walk, and read_file all add the record's LBA to the resolved origin rather than reading it absolutely. So unlike EMU3, this backend needed no code change — the test proves the existing behaviour.

Confirmed the test genuinely discriminates: temporarily dropped the origin in read_file (image.read(entry.start_block * SECTOR_SIZE, ...)), and the listing still passed while the read-back failed on pregap zeros; reverted, and it passes. Source is unchanged in this PR.

Docs

docs/formats/iso9660.md now records the LBAs-relative-to-origin assumption and flags the one untreated case as a watch-for: a real hybrid disc whose ISO LBAs count from the physical disc start rather than the track would double-count under origin + and needs its own treatment and fixture. None is known in the collection, so it is a caveat, not a case handled today.

Scope

Test + docs only (source untouched). Synthetic fixtures only (ADR-0008). Branched off main and touches only ISO 9660 files, so there is no overlap with the open #49.

Verify

• uv run ruff check . — clean
• uv run ruff format --check . — clean
• uv run pytest -q — 340 passed, 112 skipped
• uv tool install --editable . && samplerdisc --version — 0.3.0

🤖 Generated with Claude Code