Previously we used anyhow::Errors to propagate an error when we hit the
scan budget limit. This was being collapsed into a multierror which then
wasn't properly handled
[here](https://github.com/MystenLabs/sui/blob/bb17ee0d19ab1caaedd7c465ee7a0f4abf46c61d/crates/sui-rpc-api/src/grpc/v2alpha/ledger_service/bitmap_scan.rs?plain=1#L300-L306)
when downcasting, so when a query with multiple leafs that all hit the
scan limit in one round failed, it would be reported as an internal
error instead of a scan limit.

This refactor replaces the anyhow:Error with explicit error types to
make this a little harder to get wrong, and also fixes the bug.

We still use anyhow for propagating errors from the database.