Warp sync downloads runtime twice on every cold start

[replghost]

Summary

During warp sync on a live network, smoldot downloads the parachain/relay runtime (:code, ~1.5-2.5 MiB) twice on nearly every cold start. This roughly doubles cold-start time.

Root cause

In lib/src/sync/warp_sync.rs:1710, when a warp sync fragment is verified, runtime_download is unconditionally reset to RuntimeDownload::NotStarted:

self.inner.runtime_download = RuntimeDownload::NotStarted {
    hint_doesnt_match: false,
};

On a live network, GrandPa commit messages arrive every few seconds. When one arrives during the first runtime download cycle (advancing the finalized height by even 1 block), the warp sync state machine:

1. Issues a new fragment request for the updated height
2. Verifies the new fragment (1 block higher)
3. Resets runtime_download to NotStarted — discarding the already-downloaded runtime
4. Downloads :code again (~1.5-2.5 MiB)
5. Compiles the runtime again

The second download is redundant — the runtime code hasn't changed between two consecutive blocks.

Measured impact

Tested on Paseo with a fresh lightSyncState checkpoint:

	Storage proof downloads	Runtime builds	Cold start
Before fix	2 (both :code)	2	~6-8s
After fix	1	1	~2.7s

Proposed fix

Check the verified fragment header's digest for a RuntimeEnvironmentUpdated item before resetting runtime_download. This digest is the authoritative signal that the runtime code has changed. Normal blocks (vast majority) do not contain it and can safely reuse the already-downloaded runtime.

See #3198 for the implementation.