Clean up printing of testset start

[Keno]

This has bothered me for a while, but looks like it didn't bother
anyone else enough to clean it up, so here I go :). Printing when
a test starts is useful when looking through logs because you can
correlate errors on a given worker with the specific test it was
running. However, it doesn't have to be so ugly and break the nice
regular formatting of the rest of the test.

Before:

Test  (Worker) | Time (s) | GC (s) | GC % | Alloc (MB) | RSS (MB)
      From worker 3:	running testset int...
      From worker 2:	running testset math...
int        (3) |     9.51 |   0.21 |  2.2 |     507.51 |   232.42
math       (2) |    81.06 |   0.75 |  0.9 |    3391.19 |   323.18

Test Summary: |    Pass    Total
  Overall     | 1637318  1637318
    SUCCESS

After:

Test  (Worker) | Time (s) | GC (s) | GC % | Alloc (MB) | RSS (MB)
int        (3) |        started at 2019-09-27T18:35:02.761
math       (2) |        started at 2019-09-27T18:35:02.883
int        (3) |    11.91 |   0.28 |  2.4 |     723.85 |   278.07
math       (2) |    40.51 |   0.87 |  2.1 |    3287.85 |   325.49

Test Summary: |    Pass    Total
  Overall     | 2030534  2030534
    SUCCESS

[vtjnash]

seems slightly silly to have the master ask the worker to ask the master to print this information—perhaps just print it on the master in the first place?

Also, why do we have a custom lock in this file? can it just use the advisory IO lock on stdout with do-block syntax?

[Keno]

seems slightly silly to have the master ask the worker to ask the master to print this information—perhaps just print it on the master in the first place?

I figured this way would catch if the worker had died after reporting the previous test completion. Not a huge benefit though, I agree.

[Keno]

Merging, since this seems like a strict improvement. We can do refactoring of the I/O locking here at a later point.