Split out ci and lint shells, and cache nix fetcher alongside store

Does this provide any tangible benefits besides being conceptually cleaner?

Does this provide any tangible benefits besides being conceptually cleaner?

There's three changes: ci and lint split: conceptually cleaner. Also reduces the closure sizes and the network io needed. Edit: just to be clear, this will mostly benefit cold caches(like the zig 0.17 branch, and nix lock updates)

ci: cache the Nix fetcher cache alongside the store I'm working on benching the before/after. But with this cached, we will avoid some work.

I'll comment the findings once the second CI run is completed.

Just to track the numbers for the closure sizes x86_64: before: ~7.5 GiB, 735 paths lint: ~680 MiB, 118 paths ci: ~2.7 GiB, 529 paths

macos: before: ~4.5 GiB, 456 paths lint: 831 MiB, 106 paths ci: ~2.7 GiB, 347 paths

Seems like I'm hitting some ttl based cache evictions. nix gets evicted, zig-overlay stays cached. I'll mark this as draft as it's getting late here and I'll have to see if it's worthwhile removing the repeated unpacking.