Allow multi-threaded release builds and less aggressive LTO

This is fine as the possible optimisations with a single thread
are nothing we benefit from, or know we benefit from.

So let's prefer lower latency over imaginary performance improvements.

Further, LTO *should* be worth it ("thin") but not at all cost (true/"fat"),
so let's dial that in as well.

Author: Byron

.github/workflows/publish.yaml

@@ -187,6 +187,7 @@ jobs:
             gir1.2-webkit2-4.1=2.44.0-2;
 
       - uses: actions/download-artifact@v5
+        name: Download SvelteKit build output
         with:
           name: sveltekit-build
           path: ./apps/desktop/build/

Cargo.toml

@@ -100,8 +100,7 @@ rmcp = { version = "0.1.5" }
 serde_json_lenient = "0.2.3"
 
 [profile.release]
-codegen-units = 1 # Compile crates one after another so the compiler can optimize better
-lto = true        # Enables link to optimizations
+lto = "thin"      # Enables link to optimizations, but don't go all out on compile time.
 opt-level = "s"   # Optimize for binary size
 
 [profile.bench]