[CI rehearsal] measure hosted-runner costs for the flawd mutation gate

[fohara]

Measurement probe for the upcoming mutation-testing gate (dogfooding flawd ci, tracked in the flawd repo's dogfood-lash-ci effort). Times four things on ubuntu-latest: flawd install from the public release mirror (with checksum), the unentitled failure mode (expected exit 5), workspace coverage generation with cargo-llvm-cov, and the flawd Dockerfile image build. Not for merge — it will be closed once the numbers are recorded in the real mutation.yml.

[fohara]

First probe run (33015348586, ubuntu-latest, cold):

Step	Time
flawd install from the release mirror (checksum verified)	2s
Unentitled flawd ci refusal	exit 5, instant
cargo llvm-cov --workspace incl. toolchain component + cargo-llvm-cov install	114s
cargo test --workspace (separate profile, effectively cold)	151s
docker build -f Dockerfile	164s — not trusted: the Dockerfile masks cargo build failures with `

A second run with a cargo/target cache is queued to measure the warm-runner story. Cold native-toolchain setup lands around 4.5 min, which points at native execution for hosted runners; the docker-vs-native call gets made on entitled comparative runs.

[fohara]

Warm-cache run (33019217068) with an actions/cache cargo+target cache:

Step	Cold	Warm
cache restore	—	37s
cargo llvm-cov --workspace	114s	61s
cargo test --workspace	151s	85s
docker image build	164s	157s (no layer cache on hosted runners; also still masked by `

Conclusion recorded in flawd's ci-integration docs: native isolation for hosted runners (~3 min warm setup before mutant execution), Docker for local/self-hosted/compose. Measurements are done — closing this probe PR; the branch stays for re-dispatch until the real mutation workflow lands.