fix(models): tighten transient retry classification

[ehz0ah]

Description

Prevent the shared model retry classifier from treating unrelated numeric values as HTTP status codes. Also recognize common Node transport errno strings as transient failures.

The current classifier treats messages such as max_tokens=500 as transient server errors. This can repeat a deterministic invalid request. It also misses ECONNREFUSED, ECONNRESET, and ETIMEDOUT messages forwarded by Node-based gateways.

Human Involvement

• A human participated in the implementation or review loop

Related Issue

• External evidence: isRetryableAssistantError: bare status digits false-positive; Node errno forms missed earendil-works/pi#10941
• Related OpenViking false-classification history: fix(embedding): classify_api_error misclassifies 429 errors as INPUT_TOO_LARGE when request ID contains numeric pattern substrings #2369 and fix(embedding): use word-boundary matching for numeric error patterns in classify_api_error #2370

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Refactoring (no functional changes)
• Performance improvement
• Test update

Changes Made

• Require HTTP or error context before a numeric status pattern matches an error message.
• Preserve explicit forms such as HTTP 500, status code: 503, and 500 Internal Server Error.
• Classify ECONNREFUSED, ECONNRESET, and ETIMEDOUT as transient errors.
• Add classifier and real retry-loop regression tests for false-positive and transport cases.

Testing

• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have tested this on the following platforms:
  • Linux
  • macOS
  • Windows

Commands and results:

• pytest tests/unit/test_model_retry.py: 82 passed.
• ruff format --check openviking/utils/model_retry.py tests/unit/test_model_retry.py: passed.
• ruff check openviking/utils/model_retry.py tests/unit/test_model_retry.py: passed.
• Bounded mypy check for the two changed files: passed.
• A source-pinned 20-case comparison reduced mismatches from seven to the two out-of-scope cases noted below, with zero in-scope mismatches.
• tests/utils/test_circuit_breaker.py: 16 passed and two stale 401/403 expectation failures. The same two failures reproduce on unchanged upstream/main.

Checklist

• My code follows the project's coding style
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

Screenshots (if applicable)

Not applicable.

Additional Notes

This PR intentionally does not add structured exception-status handling or built-in timeout classification. PR #6047 and PR #5927 already own those overlapping changes.