fix: first try/guess fixing race of requests

[skoegl]

This PR is a draft - not tested yet...

• configure the session to use more connections allowed in connection pool determined by core count
• change simple assert check to full try except with better insight whats going on on low level errors on heavy load

[sveneberth]

I tested it with the following config:

from viur.datastore.transport import _http_internal
from requests.adapters import HTTPAdapter, DEFAULT_POOLSIZE
from multiprocessing import cpu_count

_http_internal.mount(
    "https://",
    HTTPAdapter(
        pool_connections=max(DEFAULT_POOLSIZE, cpu_count() * 4 * 2),
        pool_maxsize=max(DEFAULT_POOLSIZE, cpu_count() * 4 * cpu_count()),
        max_retries=3,
    )
)

In the cloud with the F2 instance this is cpu_count=2 and therefore pool_connections=16 and pool_maxsize=16.

Google has the following settings: https://github.com/googleapis/google-auth-library-python/blob/8cfc91db0861bc92374d708140656fb38e003ef6/google/auth/transport/requests.py#L409-L410
And requests itself has the default poolsize of 10: https://github.com/requests/requests-docs-de/blob/6e58fc50174cf463866081835da253fb37c318e4/requests/adapters.py#L65

We should therefore not go below these values in any case. But I haven't noticed any noticeable difference. So you don't have to use it as my configuration.

But I find the part with the assert the most interesting anyway.