Issue

• If publishing docs with a notebook, sphinx + myst-nb will usually default to a python3 lexer on the entire notebook unless the writer specifically set one in the notebook metadata (which frankly never happens)
• The python3 lexer can not handle cells that starts with ! such as !pip install ... or other command cells.
• This causes sphinx to throw a warning, making the CI fail because we've set Readthedocs to fail on warning
• To silence the warning, publishers must manually set the notebook lexer to ipython3 that wraps around python3 and is able to parse command cells
• This adds unnecessary overhead for publishers unfamiliar with Sphinx. They should only worry about their content, not what a lexer is

Ideally, sphinx should be able to pick a lexer for each cell but it only supports notebook-level lexer

Solution

This PR automatically adds a ipython3 lexer to any notebook listed in any of the glob patterns in ipython3_lexer_patterns. For more precision, they can also exclude files with other glob patterns in ipython3_lexer_exclude_patterns. This ensures:

• Writers can build their notebook the way they want, using any type of cells (command, magic etc)
• Flexibility in selecting which notebook are selected with an include/exclude pattern
• Re-use sphinx compiler to keep it consistent with how sphinx uses glob pattern (for exclude_patterns for example)