Add a new `reuse_validation` parameter to control validation strictness
when reusing data from a previous run (`cleanup=False`).

This addresses the issue where users have input objects with broken
`__eq__` implementations that return incorrect results, preventing them
from using the data reuse functionality.

Three validation modes:
- "auto" (default): Warn if equality comparison fails, proceed anyway
  (maintains backward compatibility)
- "strict": Raise error if equality comparison fails
- "skip": Skip input/default validation entirely (for broken __eq__)

Note: Shapes and MapSpecs are always validated regardless of mode.

Changes:
- Added `reuse_validation` parameter to Pipeline.map() and map_async()
- Updated all internal run functions (run_map, run_map_eager, etc.)
- Implemented validation logic in _compare_to_previous_run_info()
- Added comprehensive test suite (8 tests, all passing)
- Updated documentation for all affected functions