• Update tinyusb to latest, to pick up changes and also to pick up removal of all submodules (tinyusb submodules are now fetched with a script, so recursive fetch is no longer an issue). I did simple smoke tests on SAMD51, RP2040, STM, and nRF to test the update.
• Removed no-longer-used tools/usb_descriptor submodule.
• Change make fetch-submodules to use git submodule update --init --filter=blob:none. This does a partial clone of all submodules: all the tree metadata is fetch, but data blobs are fetch only on demand. This is slightly faster and does not have the idiosyncrasies of a shallow clone: non-HEAD checkouts are seamless, all tags are available, etc. Fixes Consider using git "partial clone" #7225.

Note that the huge broadcom/firmware submodule has shallow = true in .gitmodules, but it never seemed to actually work and force a shallow fetch. I did not remove that.

Here is the time difference between various ways of doing the equivalent of make fetch-submodule. Plain blobless is fastest. (EDIT to include all possibilities). Each one was done on a freshly cloned repo.

$ time make fetch-submodules # old way
...
real	4m7.562s
user	1m7.378s
sys	0m18.480s

$ time git submodule update --init --filter=blob:none
...
real	3m26.823s
user	0m48.476s
sys	0m19.834s

$ time git submodule update --init --filter=tree:0
...
real	3m51.118s
user	0m48.327s
sys	0m19.963s

$ time git submodule update --init --filter=blob:none --filter=tree:0
...
real	4m0.256s
user	0m49.165s
sys	0m19.979s