fix(empty_blocks_sanitizer): handle nil block result, flag for refetch

[rswanson]

Summary

• Equivalent of master PR fix: handle nil block result in EmptyBlocksSanitizer #19, applied directly to `signet-main` so the `publish-init4-docker.yml` workflow rebuilds `ghcr.io/init4tech/blockscout:latest` with the fix.
• `EmptyBlocksSanitizer.classify_blocks_from_result/1` crashes with `BadMapError` when the upstream RPC returns `result: null` (pruned or reorged blocks). Each crash trips `restart: :permanent` supervisor restart intensity, bringing down the `indexer` application and the whole BEAM VM → K8s crashloop.
• Adds a `%{id: _, result: nil}` match clause that skips without crashing, plus caller-side reconciliation that flags the missing block numbers with `Block.set_refetch_needed/1` so they exit the sanitizer's query set (`consensus_blocks_with_nil_is_empty_query` filters on `refetch_needed == false`) and stop being re-queried every 10s.
• Regression test mocks a nil batch response and uses `wait_for_results` to assert `refetch_needed == true`; the un-fixed code would crash and the assertion would time out.

Merge resolution notes

`signet-main` already had a defensive `Map.get(block, "transactions") || []` + `block_fields/2` signature for the case where `block` is a map but the `transactions` field is missing. That handles a different edge case than `result: nil`, so both guards are preserved in this patch.

Why this branch, not master

Master PR #19 was merged, but the daily `sync-upstream.yml` will `git reset --hard upstream/master` on init4tech/master and wipe that merge. `signet-main` is the deploy branch that actually triggers `publish-init4-docker.yml`, so the fix has to land here.

Test plan

• CI: `Indexer Tests` passes the new `skips/marks nil block` test
• Merge → `publish-init4-docker.yml` runs → arm64+amd64 build + `:latest` manifest push
• Restart the blockscout pod; confirm it stops crashlooping

🤖 Generated with Claude Code