[v2] Properly retry on hiccups on the SOCKS server

[LukeShu]

Description

@esmet reports that he sees AES CI flakes where a hiccup in the connection to the cluster causes the whole connector to quit. It should have retries! Well, it turns out that while most of the things had retries, the /connector/server-socks goroutine didn't retry, and so it'd error out, and cause all the rest of it to shut down.

As usual, I suggest a commit-by-commit review:

• The first 3 commits are doing prep-work for the fix, with an eye for auditing uses of tm.sshPort to make sure that all users of it properly retry.
• The 4th commit is the actual fix.
• The final 2 commits are generic cleanup tasks that are now-trivial and now-obvious because of the changes made above.

Checklist

• I made sure to update ./CHANGELOG.md. - yes
• I made sure to either submit a docs PR, or tell Matt about the necessary documentation changes. - no applicable changes
• My change is adequately tested. - no
• I updated DEVELOPING.md with any any special dev tricks I had to use to work on this code efficiently. - no tricks

[thallgren]

LGTM!