Fix precision loss in large integral string conversions (#3405)

## Summary

Fixes precision loss when converting large integral strings in two
runtime paths:

- `StringLiteral.to(IntegerType/LongType)`
- `partition_to_py(...)` for integral and time-based partition values
backed by integers

## Root cause

Both paths were converting through `float` before converting to `int`,
which loses precision for values outside the IEEE-754 exact integer
range.

That caused valid 64-bit integers like `LongType.max` and
`9007199254740993` to be corrupted.

## What changed

- Replaced `int(float(...))` with exact integer parsing in
`partition_to_py`
- For `StringLiteral.to(IntegerType/LongType)`, exact integral strings
now use exact integer parsing while fractional numeric strings retain
the existing truncation behavior
- Added regression tests for `LongType.max` and `9007199254740993`

## Validation

- `uv run pytest tests/expressions/test_literals.py
tests/test_conversions.py`

Closes #3404.