phd-runner is very sensitive to network issues

in #941 i saw two failures of phd-run-migrate-from-base, they were something like this:

	phd-runner: [TEST - EVENT] Error obtaining artifact from source
    e = reqwest::Error { kind: Request, url: "https://oxide-omicron-build.s3.amazonaws.com/alpine.iso", source: hyper_util::client::legacy::Error(Connect, Ssl(Error { code: ErrorCode(5), cause: Some(Io(Os { code: 131, kind: ConnectionReset, message: "Connection reset by peer" })) }, X509VerifyResult { code: 0, error: "ok" })) }
    file = phd-tests/framework/src/artifacts/store.rs
    line = 627
    path = phd_tests::migrate::from_base::migration_from_base_and_back
    target = phd_framework::artifacts::store
    uri = https://oxide-omicron-build.s3.amazonaws.com/alpine.iso

our approach to downloading then is, per test: try to fetch the artifact once if it is not already accounted for. so in the migration-from-base-and-back case above, we failed the first attempt to get alpine.iso, failed that test, retried on the next test, succeeded, and the rest passed.

it's probably worth trying a second time on a reset, and/or just getting artifacts up front before running tests, or .. something? It's also just not great that we see resets from S3 anyway.

Start at phd-tests/framework/src/artifacts/store.rs lines 623-637 and inspect phd_tests::migrate::from_base::migration_from_base_and_back. Reproduce or observe the S3 artifact fetch failure, then evaluate the retry or prefetch approaches described in the issue; done means transient connection resets no longer cause an unexpected test failure.