FIX: slow `read_off` method

[safesintesi]

Had to use this method for a project and it was incredibly slow.

Fixed changing the engine and tweaking the parameters so that no new import is required.

Also fixes:

• comments in file
• ignores edges that were previously read as faces
• direct type casting instead of casting after creating the DataFrame
• coercing error on obj.py
• changed jit decorator to njit
• little spelling mistake :)

I tried to modify the least I could.

Problems not solved:

• OFF files actually can have heterogeneous shapes... :(

[safesintesi]

Not sure how to act if a file exists and respects the standard but has no points and no shapes. At first I added a check that raises a ValueError, but then I commented it out because returning an empty point cloud might be better. Thinking about it, maybe a loop that loads a bunch of files shouldn't break just because one of them is empty. Thoughts about this?

[daavoo]

Not sure how to act if a file exists and respects the standard but has no points and no shapes. At first I added a check that raises a ValueError, but then I commented it out because returning an empty point cloud might be better. Thinking about it, maybe a loop that loads a bunch of files shouldn't break just because one of them is empty. Thoughts about this?

I think it is ok (and would prefer it) to raise an exception. People could handle exceptions in their loop

[safesintesi]

@daavoo done 🦤

[safesintesi]

Kindly pinging to remind someone about this <3

[tflidd]

Just on the commit f9bbff7cde3c996b70045dcc729d97c1d64c82a4, this works for me :+1: