net: deadlock in TestDialParallelSpuriousConnection on darwin-arm*

[bcmills]

--- FAIL: TestDialParallelSpuriousConnection (1.49s)
    dial_test.go:456: got read tcp6 [::1]:59891->[::1]:59893: i/o timeout; want EOF
FAIL
FAIL	net	4.495s

2020-03-10T00:24:30-08dee51/darwin-arm64-corellium
2020-03-02T15:39:23-12d02e7/darwin-arm64-corellium
2020-02-29T17:02:40-74f8983/darwin-arm64-corellium
2020-02-24T16:39:52-3093959/darwin-arm64-corellium
2020-01-31T20:18:54-1b7fefc/darwin-arm64-corellium
2020-01-23T21:01:12-ace25f8/darwin-arm-mg912baios
2020-01-19T14:04:09-8e0be05/darwin-arm-mg912baios
2019-10-30T00:41:31-47efbf0/darwin-arm-mg912baios

Forked from #34495.

CC @mikioh @bradfitz @ianlancetaylor @Zenly

[bcmills]

I think everything, but especially the network, is just slow on this builder. (Compare #37322, #35692, #35498, #34837.)

(CC @toothrot @cagedmantis @andybons @dmitshur, since they may suggest a builder-based solution to this cluster of flakes.)

[gopherbot]

Change https://golang.org/cl/222959 mentions this issue: net: use t.Deadline instead of an arbitrary read deadline in TestDialParallelSpuriousConnection

[bcmills]

Looks like it's a true deadlock, not just a flaky test. With the new deadline logic, the test is now hanging for nearly a full 3 minutes:

2020-04-09T01:14:53-7a4247c/darwin-arm64-corellium
2020-04-08T18:37:36-8b4cbcc/darwin-arm64-corellium
2020-04-03T14:12:25-e18a5ba/darwin-arm64-corellium
2020-04-02T15:05:15-1dbcbcf/darwin-arm64-corellium
2020-03-23T20:31:13-5b897ec/darwin-arm64-corellium

[bcmills]

The apparent deadlock in net suggests a possible connection to #34837. (Perhaps a platform bug on Corellium?)