Handle out of disk space failures better

[whyrusleeping]

The gateways have now gotten into bad states from running out of disk space a few times. Fetching content from gateways that are out of space results in something like this: https://a.uguu.se/UnPPwtbEJHS1.png

We should:

• Catch and return these errors if we encounter them.
  • Randomly closing the connection on users isnt good behaviour
• Make sure that garbage collections are run frequently enough
  • should also trigger a garbage collection if an 'out of disk space' error is encountered
• add tests for these conditions using either sized subvolumes (btrfs, zfs) or fixed size tmpfs mounts.

And while we're at it:

• Respect set 'repo size limitations'

[whyrusleeping]

Another issue that just happened and is related:

Node ran out of disk space and seemed to crash, docker attempted to restart it and failed because it needed disk space on startup and that broke it.

[whyrusleeping]

Another observation: It looks like if a node runs out of disk space and doesnt crash, it gets into a bad state where even if space is freed up, it thinks its out of space still.