Fix getLogs query (avoid repeated execution, separate tx, and improve performance)

[telackey]

Fix for #248 and #249.

This does a few things:

1. Support for "earliest", "latest", etc. block specifiers. This did not work before.

2. The default "fromBlock" is now "latest" to match geth.

3. The old code did a loop across fromBlock:toBlock, executing a query on each one inside a big tx (!). This does a single query, with no separate tx.

4. Add a max range limit (default 500) with an error message if the range is too broad (I used the same error message as cloudflare)

5. Simplify the query a good bit. This improved performance significantly in my manual testing adapted to v4. The old version, adapted to use a small range of 50 blocks took over 20s. The new version took about 700ms. A query across 500 blocks with 150K results takes less than 4s.

Some additional test cases added here: https://git.vdb.to/cerc-io/system-tests/src/branch/main/test/test_logs.py

[i-norden]

LGTM once the config is added. 100 might be low (I imagine geth doesn't timeout at that range size). And while we are at it we should add the missing canonical_header_hash condition on the getLogs query.

[telackey]

I rewrote the query to select the canonical hash of the blocks in specified range, then check for inclusion in that set. I also simplified the query structure a good bit. I had to use a v4 database for comparison, but using a small block range of 50 was running about 23s on the old query, 700ms on the new.

[telackey]

I also upped the limit to 500, as that is still decently quick for me (3 to 4s) on blocks 17,000,000 to 17,000,500 with about 150,727 results (not that I'm sure we would want to send that many...)