ENH: Make pd.to_datetime with format parameter more robust to dirty data (#62840)

[jakechirsch]

• Closes ENH: Make pd.to_datetime with format parameter more robust to dirty data #62840

This PR adds a new threshold argument to to_datetime that allows users to specify the minimum fraction of valid datetime components required for parsing to succeed. For this feature, "successful" parsing means that the function returns either a valid Timestamp or NaT (i.e., it does not raise an exception). The threshold determines whether partially-invalid values produce NaT or raise an error. This enables more flexible and robust parsing behavior for partially-invalid dates while preserving strict behavior by default (threshold=1.0).

Summary of changes

• Added threshold argument to to_datetime.
• Implemented validation logic and clamping of threshold values to [0.0, 1.0].
• Updated parsing internals to compute the fraction of valid components.
• Added tests for valid, invalid, and boundary threshold behavior.
• Added documentation: explanation, parameter description, and example.
• Added type annotations across all new argument signatures.
• Ensured all code checks and pre-commit hooks pass.

[jbrockmendel]

I don't think this is something we want to support.

[rhshadrach]

Thanks for the PR. This does not seem to address the linked issue which is about having a tolerance based on a percent of all records, but rather implements a tolerance for components within a single record.

I'm negative on this change. I do not see the value on having pandas raise vs give NaT based on the number of invalid components.