…block_pointer_from_number

Both methods previously always made an eth_getBlockByNumber RPC call.
is_on_main_chain is called every reconciliation cycle for subgraphs that
are beyond the reorg threshold, so with many syncing subgraphs this
generated a flood of unnecessary calls for blocks already in the cache.

Both methods now check chain_store.block_ptrs_by_numbers first and only
fall back to RPC when the cache has no entry or an ambiguous result
(multiple blocks at the same number due to a recorded reorg).

…_rpc

fetch_full_block_with_rpc previously called adapter.block_by_hash()
directly, bypassing the block cache and always making an eth_getBlockByHash
RPC call. It is called from ancestor_block when the cached block has no
receipts, or after walking back the chain via parent_ptr (which populates
the cache). In both cases the block may already be available in the cache.

Change it to delegate the light block fetch to fetch_light_block_with_rpc,
which goes through adapter.load_blocks() and chain_store.blocks() —
checking recent_blocks_cache and the DB before falling back to RPC.