[lwshang]

Description

Two changes to how the asset canister commits a proposed batch.

1. Commit a proposed batch in a single message. Since #4450, commit_proposed_batch awaits a self-call once it passes 35B instructions, and sets certified_data only after the last operation. If an operation fails after such an await, only the failing message rolls back: the operations applied before it stay live, and the certified tree no longer matches certified_data until the next successful commit. commit_proposed_batch now runs the state machine to completion within one message, so any failure, including running out of instructions, rolls back the whole commit, as before 0.30.2. The cost is that a proposed batch is again bounded by the per-message instruction limit. commit_batch keeps the self-call extension.

This changes ic_certified_assets::commit_proposed_batch from async fn to fn (breaking for direct callers of the crate); the Candid interface is unchanged.

2. Discard computed evidence when a batch's arguments change. propose_commit_batch now clears any evidence previously computed for the batch, and commit_proposed_batch clears it when it takes the proposed arguments, so evidence always describes the arguments the batch currently holds and new arguments need a fresh compute_evidence.

How Has This Been Tested?

• New unit test evidence_does_not_carry_over_to_reproposed_arguments covers change 2 (fails without it).
• Change 1 is at the canister entry point and only differs from the previous behaviour once a commit passes 35B instructions, which unit tests can't reach; no automated test covers it.
• All 111 ic-certified-assets lib tests pass; clippy is clean.
• src/distributed/assetstorage.wasm.gz rebuilt with ./scripts/update-frontend-canister.sh --release-build.

Checklist:

• The title of this PR complies with Conventional Commits.
• I have edited the CHANGELOG accordingly.
• I have made corresponding changes to the documentation.

🤖 Generated with Claude Code