A write refused on a halted ledger should say the index is halted #2100

When the indexer has halted, writes eventually fail with the same "novelty at max" error a merely busy ledger gives, so a client keeps retrying something that will never succeed. #2044 logs the halt's cause where those refusals happen, but it leaves the client-facing error unchanged on purpose.

This needs a decision before any code:

• the error variant and HTTP status (and whether it is retryable);
• the lockstep change in solo, whose transact classifier matches on NoveltyAtMax today.

Follow-up: #2044