[TALLEC-Scott]

Fixes #207.

@iarna/toml 2.2.5 rejects heterogeneous arrays used by PEP 735 dependency groups, for example:

[dependency-groups]
dev = [{include-group = "test"}, "ml_stack_document_extraction"]

Upgrade the existing parser to 3.0.0 in both packages and update both lockfiles. Version 3 supports these mixed arrays while preserving large integer parsing and the existing runtime/compiler requirements.

This replaces the initially proposed smol-toml migration: that dependency raises runtime requirements and requires additional compiler compatibility work. No Node support or TypeScript version change is included here.

Version 3 rejects CRLF inside comments, so normalize CRLF to LF at all three TOML parse boundaries. Do not strip lone CR characters. This is a targeted PEP 735 compatibility repair, not a claim of complete current TOML conformance.

Verification

• Mixed arrays and a valid signed 64-bit integer in an unrelated tool section no longer prevent metadata or configuration loading.
• Real indexing checks LF/CRLF, full-line/inline comments, project metadata, [tool.scip] precedence over [tool.pyright], and imported helper references.
• Production/development builds, all 21 SCIP snapshots, all 7 SCIP Jest tests, and formatting passed locally on Linux with Node 24.4.1 and Python 3.10.22.
• Internal TypeScript 4.4.4 build and all 16 upstream configuration tests passed. Both package locks remain stable after regeneration.
• Actual Node 12.0.0 and 16.0.0 parser probes passed; actual Node 16 CLI indexing verified CRLF comments, metadata/configuration and helper references.
• Native Windows/macOS execution and the entire upstream Pyright test suite were not run.