Test: Revert "Refactor connection.rs to make fail_with errors impossible"

[teor2345]

Reverts #1721

Motivation

This revert fixes the hangs in #1801 in my local tests.

Follow Up

• Re-apply the minimal fix to Panic: called fail_with on already-failed connection state, from msg=GetData #1599 (in Stop sending blocks and transactions on error #1818)
• Create a revised issue like Refactor the peer connection code to avoid failure handling bugs #1610, but excluding Panic: called fail_with on already-failed connection state, from msg=GetData #1599, and done in stages (in Re-Apply "Refactor connection.rs to make fail_with errors impossible" (double revert) #1817)

[teor2345]

Testnet took a few hours to sync, mainnet is about 60% there and making steady progress. So it seems like this PR is causing the regression in #1801.

That's sad, because I really liked this refactor.

[teor2345]

The sync finished in about 4 hours, and it only required 3 block locators to sync to tip.
(Before this revert, there was an error in almost every extend_tips.)

$ cat ~/.cache/zebra-log/zebrad-mainnet-tmp-2021-02-24-14-26-29.log | grep 'created block locator' | cut -d' ' -f1-12 | sed 's/:sync:obtain_tips:state: zebra_state::util://'

Feb 24 14:26:49.649  INFO {zebrad="d5fe566d" net="Main"} created block locator tip_height=Height(0)
Feb 24 14:56:23.928  INFO {zebrad="d5fe566d" net="Main"} created block locator tip_height=Height(106703)
Feb 24 15:00:21.611  INFO {zebrad="d5fe566d" net="Main"} created block locator tip_height=Height(108873)
Feb 24 18:48:21.532  INFO {zebrad="d5fe566d" net="Main"} created block locator tip_height=Height(1158491)
Feb 24 18:49:28.536  INFO {zebrad="d5fe566d" net="Main"} created block locator tip_height=Height(1158493)
...

[teor2345]

Closes #1801