Pandas: Support `datetime64[ns]` dataframe index as designated timestamp.

It's common to use a datetime64[ns] df.index in Pandas when dealing with timeseries.
In such case our API should just be:

buffer.dataframe(df, table_name="some_name")

This means changing the default logic of the at argument to also accept two new singleton types:

buffer.dataframe(df, ..., at=Server)  # timestamps are set by the server -- the current default.
buffer.dataframe(df, ..., at=Index)  # Use the index.

The new behaviour for the at=None default would be to:

• Use at=Index logic if the index column is a datetime64,
• or use at=Server logic if the index is any other type.

Whilst technically a breaking change, the feature change is minor and is very unlikely to affect any of our users, thus this feature will not require a new major software release number.

Start at the buffer.dataframe entry point and read the current at=None and at=Server handling, then trace the Pandas index path. Done means at=Index uses a datetime64[ns] index, at=Server retains server timestamps, and at=None selects Index only for datetime64 indexes while retaining Server behavior for other index types.