to_dataframe: an integer above 2**53 next to a float is rounded by pandas, the other readers keep it exact #340

Found while writing the docs for the 26.11.1.post1 candidate (the docs branch read the code of #331 and checked the behaviour with pandas).

For a result column that holds an integer above 2**53 next to a floating point value, the readers of #331 keep the integer exact: to_columns() returns an exact Python list, and to_arrow() and to_polars() return a string column of the exact text. to_dataframe() is the exception: it hands the exact list to pandas, which infers float64 and rounds the integer. The same data therefore gives an exact value in three readers and a rounded one in the fourth.

Proposal: build such a column as an object Series in to_dataframe() (both wheels share the decoder, so the change is in one place), and add a test next to the 2**60 + 1 and 1.5 cases that #331 already has for to_columns() and to_arrow(). It is not part of 26.11.1.post1: that candidate is built and tested, and this changes a reader's output type for a rare shape. It goes into the next patch.