fix(e2e): resolve the Darwin fixture path on bash 3.2

[SaiPisey2]

Split out of #3767 as requested.

Two fixes to end-to-end-test.sh, both about running the suite on a Darwin host.

The Darwin fixture path is not resolved on bash 3.2

fixture_metrics="${fixture_metrics::-4}-${GOHOSTOS}.txt"

A negative substring length needs bash 4.2. macOS still ships bash 3.2, where that expansion fails:

$ /bin/bash -c 'v=collector/fixtures/e2e-output.txt; echo "${v::-4}"'
/bin/bash: -4: substring expression < 0

generated_metrics on the line above uses plain interpolation and is resolved correctly, so the generated file is the Darwin one while fixture_metrics still points at the Linux fixture. Running ./end-to-end-test.sh -u on a Mac therefore overwrites collector/fixtures/e2e-output.txt with Darwin output — in my case rewriting 5330 lines of it before I noticed.

${fixture_metrics%.txt} is equivalent and works on both. Verified under bash 3.2.57 and 5.3.15, and both still pass bash -n.

node_thermal_temperature_celsius is per-machine

Its sensor names and its values both depend on the hardware, so anyone regenerating the Darwin fixture on a real Mac gets the whole sensor list as a diff rather than the intended change. It is stripped along with the other non-deterministic metrics now.

This makes no difference in CI, which reports no thermal sensors at all — it only matters when the suite is run on real hardware, which is exactly what the -u instructions ask for.

Unrelated and left alone: the sed -i /pattern/d calls in that same loop need an extension argument on BSD sed, so the stripping aborts on stock macOS. It behaves the same way on master, so it is not something this PR introduces.