Speed Windows rebuilds with rust-lld and a fast profile

[Aegyip]

Summary

• Use rust-lld.exe as the Windows linker.
• Add [profile.fast]: release codegen without thin LTO.

Test plan

• cargo build --profile fast -p postillion links on Windows.
• Incremental rebuild of one crate is much faster than --release.

[WinTone01]

Two changes in one PR with quite different risk, so let me take them separately.

[profile.fast] — no objection. It is opt-in, nothing else in the tree references it, and the reasoning (thin LTO relinks the world for a one-line edit) is right. Worth a line in the README or CONTRIBUTING so it is discoverable; a profile nobody knows about does not speed anyone up.

.cargo/config.toml — needs verification before it lands. This one is not opt-in: it changes the linker for every x86_64-pc-windows-msvc build in the repo, including the windows.yml CI job that builds and tests the whole workspace on every PR. If rust-lld cannot link this binary — gpui pulls in a lot of Windows system libraries and DirectX — then Windows builds break for everyone at once, and the first person to notice will be whoever pulls next.

Your own test plan has both boxes unticked:

• cargo build --profile fast -p postillion links on Windows.
• Incremental rebuild of one crate is much faster than --release.

I cannot check those from here (no Windows machine), and CI cannot settle it either right now: windows.yml is already failing on main for an unrelated reason (interrupt_stamps_streaming_entry_aborted in crates/engine/tests/e2e.rs), so a red run would prove nothing.

Could you run cargo build --workspace on Windows with this config in place and paste the result, plus the toolchain version you used? With that, the linker half goes in as-is. If you would rather not wait, splitting [profile.fast] into its own PR gets that part merged today.