Remove CLI/app integrations and require Python 3.12+

[adrida]

Remove the installed tracer command and all built-in connections to the Tracer app. TRACER becomes a standalone library requiring Python 3.12+, with CI coverage through Python 3.14. Fitting, routing, updates, scanning, reports, HTTP serving, and local Python/JavaScript trace recording remain available through the APIs.

Changes

• Delete the CLI implementation, console entry point, app client, authentication code, app-only tests, and CLI/app guides.
• Remove app exporters and watcher options in Python and JavaScript. Stale app credentials no longer trigger uploads; custom sinks and explicitly configured generic exports remain available.
• Replace CLI/app instructions throughout the docs and generated scan reports; add a Python scan guide and watched-span conversion example.
• Require Python 3.12+, test 3.12–3.14, and add a Python 3.12 job with the minimum core dependencies. Use 3.14 for release builds and the documented sidecar.
• Allow NumPy 1.26.4 through 2.x for modern Python wheels; raise the scikit-learn/joblib floors for NumPy 2 compatibility and the optional PyTorch floor. Remove duplicated NumPy pins from extras and obsolete downgrade advice.
• Preserve training/routing algorithms and defaults. Document FitConfig.skip_candidates for a lighter candidate sweep.
• Run JavaScript tests and the TypeScript build in CI.

Compatibility

This intentionally removes command-line/app-client interfaces and drops Python 3.9–3.11. Callers must use the library APIs, remove app-specific watcher options, and upgrade older Python environments. Dependency upgrades do not guarantee compatibility with previously fitted scikit-learn artifacts; retain the fitting environment or refit when upgrading. The archived research PDF retains its historical author attribution.

Validation

• Current-head GitHub CI passed all five jobs: Python 3.12/3.13/3.14, minimum core dependencies on 3.12 (NumPy 1.26.4, scikit-learn 1.4.2, joblib 1.3.2), and JavaScript tests/build.
• Full local Python suite on 3.12.13 with NumPy 2.5.3 and scikit-learn 1.9.1: 94 passed, 3 skipped. Skips are the uninstalled optional embedding extra and two existing synthetic fits without a deployable policy.
• JavaScript: 22 tests passed, TypeScript build and npm package dry run passed during the CLI/app removal; JavaScript is unchanged by the Python update.
• Wheel and sdist built successfully; inspected for absent CLI/app packages and console entry points. Wheel metadata requires Python 3.12+, and pip dry runs correctly reject 3.11 and accept 3.12/3.14.
• All optional extras resolve to binary wheels on Linux for Python 3.12 and 3.14. This is a dependency-resolution check, not a runtime test of optional models.
• Dependency consistency (pip check), docs/metadata consistency, and git diff --check passed.