Allow plain text .qmd files as source notebooks

[bhoov]

Addresses #1461

Using quarto and its VS code extension, I find that writing .qmd files to be a smoother interactive alternative to .ipynb files. That .qmd files are plain text comes with several advantages:

1. .qmd seamlessly integrates with Cursor AI/other AI copilots.
2. .qmd is fully compatible with standard git tooling
3. .qmd works better with VIM keybindings
4. .qmd files don't need a special nbdev_clean step to remove cell metadata and outputs, meaning your source files are not altered in any way by nbdev's transpilation process (something that bothers me immensely when developing in .ipynb)

Turns out, nbdev doesn't need many changes to implement this feature.

1. Allow export globbing functions to search for .qmd in addition to .ipynb
2. Implement a read_qmd/write_qmd function for converting the .qmd to/from nbdev's AttrDict format. This means two-way sync (via nbdev_update) also works for .qmd and its corresponding .py files.
3. Because outputs are not stored inside .qmd files, I use execnb's run_all to generate outputs for the docs inside _proc/-cached .ipynb files.
4. The custom frontmatter parser needed some tweaking to allow cells to include general markdown after the custom frontmatter.

It looks like there have been other attempts to allow .qmd support for nbdev (see this quarto issue) or allow plain-text support (see #1499). However, .qmd support is still missing in the current version of nbdev, and the latter seems to introduce additional dependencies without explicitly mentioning support for .qmd files. This PR allows you to seamlessly develop using a mix of .qmd and .ipynb, whichever you prefer, with no additional dependencies.

I've written up a small tutorial for setting good VSCode defaults in nbs/tutorials/develop_in_plain_text.qmd

A few notes of caution and room for improvement:

1. Ensure that all files under nbs/ have distinct names: no 00_core.ipynb and 00_core.qmd, as both of these will create the intermediate _proc/00_core.ipynb
2. Currently, the nbdev_prepare will run executable cells in .qmd documents twice: 1x when testing and, because outputs aren't saved, 1x when generating the docs.

The PR is in a pretty stable position already (see this fork of nbdev rewritten entirely using .qmd files). There may be edge cases that I haven't considered, but in all I hope this is nearing a good shape to distribute.