Python: validate non-optional union settings - #8926

Motivation & Context

Settings declared as int | float currently retain raw environment or dotenv strings, including invalid values. This can defer configuration failures until agent execution instead of reporting the invalid setting at startup.

Description & Review Guide

• What are the major changes? Apply the existing per-arm coercion logic to every union type, using the same union-origin check already used by settings validation. Add six regression cases covering integer/floating-point conversion and rejection from both environment and dotenv sources.
• What is the impact of these changes? Valid numeric strings become the declared numeric type; invalid strings raise the existing field/source error without exposing the supplied value. No new API or dependency.
• What do you want reviewers to focus on? Reuse of existing optional-union handling and preservation of its fallback/error behavior. All six regressions fail before the fix; the 124 settings tests and full core unit suite pass (8,074 passed, 19 skipped, 2 expected failures). Source Pyright, all five test-typing checkers, Ruff lint/format, and diff checks pass.

Related Issue

Fixes #8907. No competing open PR was found.

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change.