[FEA] Document all behaviors associated with pandas_compatible mode

## Problem

Historically cudf has aimed to match pandas to the greatest extent possible, but there have always been some differences. Some of these are simply features that have not yet been implemented, while others are intentional choices typically made for performance reasons. Some of these differences are documented here:
- https://docs.rapids.ai/api/cudf/stable/user_guide/pandascompat/
- https://docs.rapids.ai/api/cudf/stable/user_guide/pandas-comparison/

One particular difference I'd like to call out is join ordering, which is described in https://github.com/rapidsai/cudf/issues/14001.

Originally, perfectly matching pandas was not the goal of cudf. With cudf.pandas, however, the importance of that goal has shifted. At minimum, now, when the `pandas_compatible` option is set, we are aiming to achieve complete compatibility with pandas, at least for APIs that are implemented in cudf. https://github.com/rapidsai/cudf/issues/18659 captures our desire to achieve this goal.

## Proposed Action

As we progress towards https://github.com/rapidsai/cudf/issues/18659, we should ensure that all of the effects of setting the `pandas_compatible` option are documented since these effectively also constitute the differences in behavior between cudf and pandas when the option is off.