Add experimental AssayScript multi-read runtime

[dnncha]

Outcome

Turns the AssayScript v2 compiler into an executable, fail-closed experimental multi-read workflow while keeping the native DotMatch commands authoritative for supported high-throughput use.

This is intentionally stacked on #54 so the runtime can be reviewed independently from the 0.2 platform and packaging work.

Runtime

• streams synchronized R1/R2/I1/I2 FASTQ or FASTQ.GZ inputs by canonical read ID
• verifies compiled source, target-library, and allowed-combination SHA-256 fingerprints
• executes fixed or anchor-relative extraction, jitter, and forward/reverse-complement orientation
• supports bounded Hamming and Levenshtein matching
• preserves segment ambiguity rather than resolving by target order
• filters biologically allowed cross-segment tuples and performs deterministic joint rescue only when one tuple remains
• requires explicit --accept-findings when the compiler recorded review findings
• writes assignments.tsv, unique counts.tsv, watch-compatible events.jsonl, and fingerprinted summary.json
• stages outputs and commits them only after successful completion

Interface

assaycode compile assay-v2.toml --out assay.plan.json
assaycode inspect assay.plan.json
assaycode execute assay.plan.json \
  --r1 R1.fastq.gz --r2 R2.fastq.gz --i1 I1.fastq.gz \
  --out run/ --accept-findings

Scientific boundary

The documentation and manuscript describe this as a Python-based experimental runtime with focused unit evidence. This PR does not claim public multi-read dataset validation, production throughput, or universal assay/platform coverage. Native DotMatch assignment remains the supported high-throughput path.

Verification

• 557 passed across python/tests
• assaycode-readiness: ready
• git diff --check

Dependency

• Base PR: Introduce AssayCode and the DotMatch 0.2.0 assay platform #54