fix: preserve string types in schema builder

[sapunyangkut]

What changed

• preserve typed JSON string values as schema strings instead of re-parsing them as booleans or numbers
• keep the existing None, native boolean, integer, float, and empty-string behavior
• raise an explicit TypeError for unsupported value types rather than creating an invalid string schema
• update focused tests to reflect the typed JSON contract and cover numeric-looking, boolean-looking, null-like, quoted-content, and empty strings

Since #4271, workflow configuration is loaded with nextflow config -o json and json.loads(), so a Nextflow value such as '3' already reaches build_schema_param() as Python str("3"). The retained legacy string branch then converted it to an integer, losing the type declared by the pipeline author.

The pre-JSON cache format used flat params.* keys, while the current reader expects nested config["params"]; those caches do not reach this conversion branch. Cache migration is therefore kept out of this focused fix.

Fixes #4367.

Validation

• targeted schema parser tests: 10 passed
• direct typed boundary assertions for strings, None, booleans, integers, floats, and unsupported lists
• Ruff 0.15.12 check and format check on both changed files
• python -m py_compile on both changed Python files
• git diff --check

The broader TestSchema fixture currently stops on this Windows checkout with a pre-existing Jinja2 TemplateNotFound for create/template_features.yml; the new setup-free regression tests and direct typed assertions pass.

PR checklist

• The change is based on and targets dev
• The change includes focused regression coverage
• The commit includes a DCO sign-off

Automation disclosure

This focused bug fix was prepared and tested with OpenAI Codex automation and independently red-teamed by a separate local agent. The contributor remains responsible for the submitted change; no human review is claimed.