Fix fatal error handling in V3 API

[simolus3]

This is related to #119.

This fixes some of the problems @osaxma has discovered and reported in that issue:

• We sometimes set the _pending operation without properly clearing it when we're done.
• When we receive a fatal error message, we sometimes continue to wait for a ReadyForQueryMessage that we'll never receive.
• The awkward setup with two completers in _sendAndWaitForQuery can lead to errors being swallowed.

This PR should fix all of them, so I've migrated the fixme test into the main v3 test file.

[simolus3]

There is some duplicate code in the error handling logic now, I'll look into ways to improve that as well.

[codecov-commenter]

Codecov Report

Patch coverage: 83.49% and project coverage change: -0.29% ⚠️

Comparison is base (18b22c5) 86.72% compared to head (c49eff8) 86.43%.
Report is 1 commits behind head on master.

❗ Your organization is not using the GitHub App Integration. As a result you may experience degraded service beginning May 15th. Please install the GitHub App Integration for your organization. Read more.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #126      +/-   ##
==========================================
- Coverage   86.72%   86.43%   -0.29%     
==========================================
  Files          29       29              
  Lines        2455     2492      +37     
==========================================
+ Hits         2129     2154      +25     
- Misses        326      338      +12     

Files Changed	Coverage Δ
lib/src/message_window.dart	100.00% <ø> (ø)
lib/src/v3/connection.dart	85.30% <83.00%> (-2.20%)	⬇️
lib/src/server_messages.dart	77.69% <100.00%> (+0.34%)	⬆️

... and 1 file with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[isoos]

I'm following maybe 50% of what happens here with the states and the code, but overall looks reasonable. Thanks for looking into it!