Use total_memory to calculate "time" Docker limit

[humitos]

4Gb servers:

• total memory: 3417
• memory limit: 2700
• time limit (using memory limit): 700
• time limit (using total memory): 900 (15 minutes)

8Gb servers:

• total memory: 7935
• memory limit: 7200
• time limit (using memory limit): 1800
• time limit (using total memory): 2000 (33 minutes)

These numbers are more similar to what we communicate in
https://docs.readthedocs.io/en/stable/builds.html. We are giving an increase of
3 extra minutes in Read the Docs for Business instead of 4 minutes less in Read
the Docs Community.

>>> round(3417 - 750, -2)
2700
>>> round(2700 * 0.25, -2)
700.0
>>> round(3417 * 0.25, -2)
900.0
>>> round(7935 - 750, -2)
7200
>>> round(7200 * 0.25, -2)
1800.0
>>> round(7935 * 0.25, -2)
2000.0

[ericholscher]

I'm still not convinced that we shouldn't just be setting this explicitly, instead of doing a bunch of magic to try and have it line up with our intended goal of an obvious set of values for specific servers. This seems fine I guess, but still over-engineered to me.

[agjohnson]

This makes more sense 👍

[agjohnson]

I don't think it helps to spend more time on reverting this logic. The original approach using static values set the wrong limits on multiple occasions -- the approach we have now has been working fine.