Fix pyright type errors and AWS Bedrock semantic tag bug

[pragyan-amp]

Summary

• Resolved all 74 pyright type errors across the codebase
• Fixed bug where AWS Bedrock LLM assigns semantic tags to numeric columns (LONG, BIGINT, etc.), causing Stitch jobs to fail with "Semantics on fields of type LONG are not supported"

Changes

Type fixes:

• Aligned ModelInfo TypedDict across LLM providers
• Fixed WizardState.models type hint
• Added proper None handling and type narrowing throughout
• Fixed except clause ordering and Callable imports

Stitch/PII fix:

• Added NUMERIC_TYPES filter in stitch_tools.py to strip semantics from numeric columns
• Updated PII detection prompt to instruct LLM not to tag numeric columns

Test plan

• All 626 unit tests pass
• Pyright reports 0 errors
• Ruff linting passes

[punit-naik-amp]

LGTM