chore: Memory benchmark workflow #2

	name: Memory benchmark

	on:
	pull_request:
	types:
	- opened
	- reopened
	- synchronize
	- labeled

	permissions:
	contents: read

	jobs:
	memory-benchmark:
	name: Memory benchmark (memray)
	runs-on: ubuntu-latest # GitHub-hosted; Docker preinstalled
	# Only runs when the PR carries the label. The 'labeled' trigger above lets
	# adding the label to an open PR start the run immediately; every other PR
	# event is a no-op for this job.
	if: contains(github.event.pull_request.labels.*.name, 'check-memory-benchmark')
	steps:
	- uses: actions/checkout@v4

	# No setup-python / pip install on the host: the Docker image
	# (python-3.12-slim-perf-Dockerfile) already has Python 3.12 + memray +
	# deps, and entrypoint.sh downloads the native lib at container start.
	# baseline.json comes from the repo. This is the exact path devs run
	# locally via `make memory-use-bench`. Exits non-zero on regression
	# (> baseline * threshold) -> the check goes red. The Makefile forwards
	# GITHUB_TOKEN and GITHUB_STEP_SUMMARY into the container.
	- name: Run memory benchmark
	env:
	GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
	run: make memory-use-bench

	# Upload all three flamegraph views per scenario (peak/leaks/temporary).
	# if: always() so the reports are available even when the run failed on a
	# regression -- exactly when they are most useful.
	- name: Upload flamegraph reports
	if: always()
	uses: actions/upload-artifact@v4
	with:
	name: memray-flamegraphs
	path: tests/perf/reports/*.html
	if-no-files-found: warn

Provide feedback