16354: storage: Delay initial metrics until system config is ready r=bdarnell a=bdarnell

This was previously done too soon at startup and would clutter the
logs (mainly in tests) with "failed initial metrics computation".

Fixes #13560

29817: client, roachpb: eliminate log spam when loadgens are killed r=andreimatei a=andreimatei

When the TPCC loadgen is CTRL-C'ed, the logs are spammed with:
I180906 22:02:41.239771 27177127 internal/client/txn.go:625  [n1] async rollback failed: TransactionStatusError: already committed (REASON_UNKNOWN): "sql txn" id=57c518aa key=/Table/61/1/1224/0 rw=true pri=0.03781170 iso=SERIALIZABLE stat=COMMITTED epo=0 ts=1536271361.068457287,0 orig=1536271361.068457287,0 max=1536271361.073978228,0 wto=false rop=false seq=11 int=5

The "async rollback" part refers to the rollback being done with a
canceled ctx (presumably a dropped connection's ctx). I believe the
error happens because there's a commit in flight when the ctx is
canceled. This patch lowers the message's level for this case.

Release note: None

29987: github-post: add timeout handling to the stress issue poster r=andreimatei a=andreimatei

Add explicit support to the issue poster for timeouts:
- on all runs, publish an artifacts file with a list of slow tests
- when timeouts happen, distinguish between the case where the test
currently running at the timeout point is the culprit (i.e. if it is the
longest running test) versus situations where that test is just an
innocent bystender

This patch also spruces up the github-post script in various ways. Among
them there's now better support for running it on an input that comes
directly from a test run, and not from a stress wrapper.

Release note: None

30006: opt: fix panic during SELECT MIN(NULL) r=rytaft a=arjunravinarayan

Fixes #29695.

Release note (bug fix): fix a crash when SELECT MIN(NULL) was run with
the optimizer enabled.

30008: gossip: allow receipt of "loopback infos" r=bdarnell a=petermattis

Receipt of loopback infos was disabled in #29398, but doing so had the
unfortunate effect of allowing gossip state to temporarily diverge
between nodes. Rather than disallowing loopback infos, we now ratchet
the gossip monotonic clock in order to avoid the assertion around the
gossip highwater timestamps.

Fixes #29992
Fixes #20986

Release note: None

Co-authored-by: Ben Darnell <ben@cockroachlabs.com>
Co-authored-by: Andrei Matei <andrei@cockroachlabs.com>
Co-authored-by: Arjun Narayan <arjun@cockroachlabs.com>
Co-authored-by: Peter Mattis <petermattis@gmail.com>

It's dead code; was just a convenient way of avoiding test log spam way
back in cockroachdb#16354.

Release note: None