New fork detection incompatible with tqdm

[bjenik]

It appears the new fork detection is very aggressive, making tensorstore incompatible with packages like tqdm, which is fairly widely used for progress tracking in the AI/ML ecosystem and causing downstream issues. I am not certain if there was actually a problem that is now successfully detected and prevented or if we are looking at what then ends up being a false positive, but I did not experience any issues using this combination until now.

Minimal example:

import tensorstore as ts
from tqdm import tqdm 

dataset = ts.open({
    'driver': 'n5',
    'kvstore': {
        'driver': 'file',
        'path': 'tmp/dataset/',
    },
}).result()

for i in tqdm(range(10)):
    print(i)

[laramiel]

Yes, there were numerous freeze/lockups that have caused issues due to calling fork. However we agree that it was too aggressive, and I submitted a change to relax the fork detection quite a bit here: f2a81b6

Unfortunately I'm waiting on a pypi resource update to do another point release:
pypi/support#5719