ci: build the release test step on 8 cores

`check/release` ran on `lto_cores` (4) because its test archive's
unbounded ThinLTO links could otherwise overlap past the runner's
20 GiB, and still peaked at 22 GiB locally. With the archive's links
bounded, the link step split from compilation and test binaries
carrying line tables, its peak no longer depends on how many links
overlap.

Measured building the step's whole chain cold on 8 cores, one
derivation at a time as CI does, summing every process of each build:

    dataplane-deps                     74 s    6.2 GiB
    dataplane-tests-deps               98 s    4.9 GiB
    dataplane (ships: full debug info) 252 s    6.7 GiB
    dataplane-cli                      26 s    1.4 GiB
    dataplane-init                     51 s    2.3 GiB
    tests.all                         650 s   12.4 GiB

`tests.all` took 929 s and peaked at 22 GiB on 4 cores before this
series. Give the step its own `release_test_cores`; the other jobs on
`lto_cores` (doctests, sanitizers, coverage, the model checkers) were
not measured with these changes and stay on 4.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Daniel Noland <daniel@githedgehog.com>

Author: daniel-noland

ci.just

@@ -47,8 +47,26 @@ export NEXTEST_TEST_THREADS := cores
 # Changing this forks nothing: `--cores` is deliberately outside the derivation hash, and
 # `nix path-info --derivation` returns the same `.drv` at 10, 4 and 1. It is a scheduling knob,
 # not a build input.
+#
+# Those numbers predate the test archive bounding its own links; see `release_test_cores`.
 lto_cores := "4"
 
+# The `check/release` test step. Its test archive now bounds how many of its ThinLTO links run at
+# once (`LINK_SLOTS` in default.nix) and carries line-table debug info, so its peak no longer
+# depends on how many links happen to overlap. Measured building the step's whole chain cold on 8
+# cores, one derivation at a time as here, summing every process of each build every two seconds:
+#
+#     dataplane-deps                     74 s    6.2 GiB
+#     dataplane-tests-deps               98 s    4.9 GiB
+#     dataplane (ships: full debug info) 252 s    6.7 GiB
+#     dataplane-cli                      26 s    1.4 GiB
+#     dataplane-init                     51 s    2.3 GiB
+#     tests.all                         650 s   12.4 GiB
+#
+# against `tests.all` alone taking 929 s and peaking at 22 GiB on `lto_cores` before. The other
+# jobs on `lto_cores` were not measured with these changes and stay there.
+release_test_cores := "8"
+
 [private]
 _lab-common := " docker_sock=/run/docker/docker.sock" + " oci_repo=ghcr.io" + " debug_justfile=" + debug_justfile
 
@@ -59,6 +77,10 @@ _lab := "jobs=" + jobs + " cores=" + cores + _lab-common
 [private]
 _lab-lto := "jobs=" + jobs + " cores=" + lto_cores + _lab-common
 
+# See `release_test_cores`.
+[private]
+_lab-release-test := "jobs=" + jobs + " cores=" + release_test_cores + _lab-common
+
 [default]
 [private]
 @default:
@@ -70,7 +92,7 @@ check-fmt:
     just {{ _lab }} fmt --check
 
 check-test profile:
-    just {{ if profile == "debug" { _lab } else { _lab-lto } }} profile={{ profile }} test
+    just {{ if profile == "debug" { _lab } else if profile == "release" { _lab-release-test } else { _lab-lto } }} profile={{ profile }} test
 
 check-clippy profile:
     just {{ _lab }} profile={{ profile }} clippy