Add EuclideanDistance.unsafe, module configuration, typeclass instances for Distance, and PixelDefinition and Pixels3D tests from looptrace

[vreuter]

TODO:

• CI passing
• CHANGELOG update
• version update (drop SNAPSHOT)

Updated base feature branch from #28

[cchlanger]

Code Review — Approving ✅

Reviewed the full diff for #29. Solid, well-tested work — clean typeclass usage, good coverage, and consistent propagation of the error-message change.

What this PR delivers

• EuclideanDistance.unsafe(Length) — a validating wrapper that throws on a negative length, complementing the existing Either-returning parse.
• New configuration module with pureconfig/scopt parsing of pixel definitions (PixelDefinition, Pixels3D).
• Show / SimpleShow instances for geometry.Distance.
• Tooling: sbt → 1.12.13, plugin bumps, and a CI Java matrix trimmed to LTS releases Temurin 17 & 21.
• Version bump to 0.6.0 + changelog.

Strengths

• Good test coverage for the new behavior: unsafe is covered for the type restriction (assertTypeError/assertCompiles), the happy path, and the failure path with an exact error-message assertion.
• Error handling in the pureconfig readers is clean — emap + leftMap into CannotConvert with value/toType/because is exactly the right pattern.
• The error-message change is consistently propagated (source + TestDistanceJsonCodec + TestDistance), so nothing silently drifts.

Should-fix / worth confirming

1. EuclideanDistance.unsafe scaladoc was misleading — "Try to interpret" read like it returned an Either/Option, but it throws IllegalArgumentException on a negative length. ✅ Fixed: reworded and added a @throws tag.
2. Show[Distance] roundtrip depends on squants' parser — the roundtrip test uses Double.MaxValue, so large values render in scientific notation; the guarantee rests on squants' parsing grammar. ✅ Documented this dependency in a comment.

Nits

1. Test-helper duplication — maxSize, genMicroOrNano, MicroOrNano, genPlusMinusMillion are duplicated between the imaging and configuration suites. Different modules, so sharing would need the testing module — probably not worth it now, just noting the drift risk.
2. minSuccessful(10000) in two property tests is heavier than needed for verifying linear scaling; 1000 would likely give the same confidence at lower CI cost.

Questions

• deriveReader[Pixels3D] over private x/y/z fields — derivation works via the product Mirror regardless of field visibility, and the test confirms it. Just confirming it's intentional that x/y/z become part of the public config API despite the fields being private.

Follow-up fixes applied after review

• Corrected the unsafe scaladoc (item 1) and added a @throws tag.
• Documented the Show[Distance] roundtrip dependency (item 2).

Verification

• Full aggregated test runs green locally (68 tests / 20 suites), including the roi and testing suites that were previously skipped when zarr dependency resolution failed.
• Confirmed the earlier CI failure's root cause was Java 19's outdated truststore (missing the Sectigo Root R46 anchor); dropping 19 resolves it, and 17/21 already trust the root, so no truststore workaround is needed.

Good to merge.