Add configurable z-score severity thresholds

[Ochibueze474]

This PR adds configurable severity thresholds as requested in issue #69.

classify_severity() now accepts optional low_threshold, medium_threshold, and high_threshold parameters, defaulting to the original values (1.0, 2.0, 3.0) so nothing changes for existing callers. detect_ml_anomalies() passes these through too, so custom thresholds work end to end.

Added tests for non-default thresholds in both test files, and updated the README table since Python now matches Node here.

While I was in these files, ruff flagged some older-style type hints (List, Optional) that predate this change, so I cleaned those up too since it was a quick fix.

All 85 tests pass, and ruff/pylint are both clean on the files I touched.

Since Node already supports this, I didn't think a follow-up issue was needed for the other SDK, let me know if you'd prefer otherwise.

Fixes #69

[AnupamKumar-1]

Should we validate that low_threshold < medium_threshold < high_threshold here? Without that, an out-of-order configuration can silently produce incorrect severities. A ValueError would make invalid configuration fail loudly; alternatively, documenting this ordering requirement would make the contract explicit.