remove kv stores from install.py

[minrk]

As discussed in #155, I don't think we should be handling installation of kv stores. I think it's very rare that installing a kv store is likely to be done via install.py, as

• ~everywhere install.py is useful, a file provider is probably better,
• ~everywhere a kv store is desirable, an official container image is more likely how these will be installed, not in the same container as the Hub
• installing go binaries isn't complex for our target users, and both tools have simple copy/paste installs

[consideRatio]

I think from a maintenance perspective, its better that this is decoupled. If for example etcd reports a critical vulnerability and requests everybody patch, suddenly we find ourself in need of making a release right? I'd like to avoid this situation and think its entirely reasonable that this is offloaded to the jupyterhub admin or user that provides a distribution.

I think this PR requires approval of at least one more person than me to be merged, and it has failing tests currently I see. But, the changes LGTM from what I can tell.

[consideRatio]

Also with 3.5.7 we keep hanging, so maybe trial 3.4.x now for the purpose of diagnosing, while we should support 3.5.7 also.

I'm on a mobile now and will not act.

[GeorgianaElena]

Downgrading etcd to 3.4.24 raised the following error in tests:
etcdmain: error verifying flags, invalid value "3.4.24" for ETCD_VERSION: parse error. See 'etcd --help'.

It looks like it's because etcd uses internally the ETCD_VERSION env var etcd-io/etcd#11225 and it's a bool. So, the last commit renames it to ETCD_DOWNLOAD_VERSION.

Let's see if this also fixes the hanging part

[minrk]

Thanks!