Don't reuse connections when StreamReader has an exception

[TimothyFitz]

In our production environment, we occasionally experience connection timeouts that "stick" in the aiomysql Pool. Every subsequent connection acquired from the pool will fail on first use.

My production environment full traceback (with private code redacted): https://gist.github.com/TimothyFitz/2d929f907c4cad47a7c8da8e0944b19e

This PR adds a test to simulate our production failure by directly calling connection_lost on a connection in the free pool.

The pool code already checks for gracefully closed connection via _reader.at_eof(). The fix is to also check for _reader.exception().

I think #132 accidentally conflated two issues, and the fix identified there is needed in addition to pool recycling: #132 (comment)

Tested on:
(production) Ubuntu 16.04.5 LTS, Python 3.5.2, aiomysql 0.0.17 + this patch
(dev laptop) Ubuntu 18.04.1 LTS, Python 3.6.5, this PR (master + patch)

[codecov]

Codecov Report

Merging #339 into master will increase coverage by 0.26%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #339      +/-   ##
==========================================
+ Coverage   92.91%   93.17%   +0.26%     
==========================================
  Files           9        9              
  Lines        1129     1129              
  Branches      161      161              
==========================================
+ Hits         1049     1052       +3     
+ Misses         56       54       -2     
+ Partials       24       23       -1

Impacted Files	Coverage Δ
aiomysql/pool.py	96.83% <100%> (+1.89%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 67b01a8...b19d816. Read the comment docs.

[terricain]

Looks good, great work :)