150 test binaries do not fit on a runner, so stop trying

150 test binaries do not fit on a runner, so stop trying Reclaiming the preinstalled toolchains takes a runner from ~14GB free to 34GB, and `cargo test --workspace --all-features` still died linking the last of the test binaries — twice, with `rustc-LLVM ERROR: IO failure on output stream: No space left on device`. Debuginfo settings were never going to close a gap that size. The cause is the shape of the suite rather than anything wasteful: 150 integration binaries, 57 of them in `dam-db` and 38 in `dam-api`, each statically linking the whole graph — sqlx, the aws sdk, tantivy, image. Sharding by crate bounds the disk a single runner has to hold instead of trimming the suite to fit, and shortens the wall clock as a side effect. `fail-fast: false`, because one shard hiding another's failure would defeat the point of separating them. Drops `dtolnay/rust-toolchain` from the job while here. Every cargo call goes through `mise exec`, so the second toolchain was a gigabyte of disk and an open question about which compiler actually ran.