tool: fix memory use in parallel mode

[icing]

The curl tool was creating a new transfer every time it checked if it needed to add one to reach max parallelism. This led to eventually all configured transfers to have easy handles created.

Limit the creation again to the ones needed for max parallelism.

scorecard.py: set --out-null only for curl versions that support it

When measuring mem use in scorecard for a non-debug curl build, notice the memory use when going from 1 parallel transfer to 6:

> python3 tests/http/scorecard.py -r --request-count=20000 --samples=3 h2
curl  Requests in parallel to httpd/2.4.67
version  size  total     1 max [cpu/rss]          6 max [cpu/rss]         25 max [cpu/rss]         50 max [cpu/rss]        100 max [cpu/rss]        300 max [cpu/rss]
8.15.0   10KB  20000  3130 r/s [50.3%/12MB]    9178 r/s [92.5%/13MB]   10827 r/s [98.3%/19MB]   10912 r/s [99.0%/26MB]   10857 r/s [99.4%/36MB]   10408 r/s [99.6%/65MB]
8.16.0   10KB  20000  3351 r/s [47.4%/12MB]   10851 r/s [94.9%/41MB]   14171 r/s [97.5%/31MB]   14657 r/s [98.8%/30MB]   15044 r/s [99.3%/27MB]   14802 r/s [99.5%/32MB]
8.17.0   10KB  20000  3391 r/s [47.2%/12MB]   10879 r/s [94.2%/42MB]   14501 r/s [98.1%/33MB]   15021 r/s [98.8%/30MB]   15337 r/s [99.3%/26MB]   15199 r/s [99.5%/31MB]
8.18.0   10KB  20000  3474 r/s [50.8%/11MB]   10394 r/s [95.6%/44MB]   13005 r/s [98.2%/34MB]   13451 r/s [98.8%/29MB]   13668 r/s [99.3%/27MB]   13701 r/s [99.6%/31MB]
8.19.0   10KB  20000  3430 r/s [50.3%/12MB]   10614 r/s [95.8%/39MB]   13099 r/s [96.2%/32MB]   13629 r/s [98.8%/28MB]   14066 r/s [99.3%/25MB]   13726 r/s [99.6%/28MB]
8.20.0   10KB  20000  3446 r/s [50.5%/11MB]   10768 r/s [95.4%/59MB]   13365 r/s [98.3%/42MB]   13749 r/s [98.8%/34MB]   14136 r/s [99.3%/30MB]   13946 r/s [99.6%/30MB]
8.21.0   10KB  20000  3395 r/s [50.9%/11MB]   10634 r/s [95.3%/57MB]   13370 r/s [98.3%/38MB]   13506 r/s [98.8%/33MB]   13908 r/s [99.3%/30MB]   13734 r/s [99.4%/30MB]
master   10KB  20000  3952 r/s [48.6%/11MB]   12184 r/s [95.1%/58MB]   15636 r/s [99.8%/44MB]   16070 r/s [99.8%/36MB]   15867 r/s [99.8%/29MB]   15841 r/s [99.9%/30MB]
PR       10KB  20000  3984 r/s [49.2%/11MB]   12711 r/s [95.1%/11MB]   16034 r/s [99.8%/12MB]   16596 r/s [99.9%/14MB]   16503 r/s [99.9%/16MB]   16303 r/s [99.9%/25MB]

[testclutch]

Analysis of PR #22277 at 4399487a:

Test 311 failed, but it has been 2.1% flaky lately, so it's probably NOT a fault of the PR. Note that this test has failed in 12 different CI jobs (the link just goes to one of them).

Generated by Testclutch