[KonScanner]

Overview

Two batching routes collapse per-row RPC into per-batch RPC for blocks,
nonces, codes, slots, balances and proxy_slots, and eleven new
datasets cover the token and proxy standards that had no dataset at all.
Five silent correctness bugs found in the existing per-row paths are fixed
alongside.

Reasoning

Most datasets read one thing per row and cost one request per row. Both new
routes are on by default and both fall back to the per-row path when an
endpoint refuses, so they change what results cost and never the results
themselves.

JSON-RPC batching puts N calls in one HTTP body. It asks nothing of the
node beyond JSON-RPC itself. The batch size negotiates itself down when a
provider caps it (OP Mainnet answers 413 above ten calls, Base returns
-32014), and the discovered size is kept for the rest of the run.

State overrides replace the target's code with a 25-byte extractor loop,
turning "read one slot" into "read every slot I name". The extractor is
PUSH0-free deliberately: eth_call at a historical block runs under that
block's fork rules, and PUSH0 only became valid at Shanghai (mainnet
17034870).

The dangerous case is an endpoint that accepts the override parameter and
ignores it, because our slot keys would then reach the real contract and its
fallback output would be decoded as storage. Three defences: an exact
32*n length check, a sentinel slot set through stateDiff and read back as
the final word, and a consecutive-failure count that stops asking an endpoint
that fails the same way three times running. stateDiff is used and never
state, because state replaces an account's whole storage and would return
a well-formed file of zeros.

eth_getStorageValues sits above both rungs where a node has it — it takes a
{contract: [slots]} map, which an injected extractor cannot do at any batch
size. Support is not cached either way, because a load-balanced pool holds a
mix of node versions.

Fixes

Reading the per-row paths closely enough to batch them turned up five bugs,
all silent:

• contracts held each hash column's value in the other column.
  code_hash is the usual join key for contract identity, so every match
  against an EXTCODEHASH found nothing rather than erroring.
• Eleven log datasets accepted --address, --from-address, --to-address
  and --topic0..3 under --txs, printed them in the run summary, and then
  ignored them. The run returned rows it had been asked to exclude.
• call_outcome recognised four of geth's execution-failure messages and
  treated the other ten as node failures, failing whole chunks where a revert
  wrote one null.
• --multicall-require-success was a no-op on a batch of one — which, since
  rows are grouped by block, is every batch for a single address over a range.
• Five calls_for_row implementations built their target with
  Address::from_slice, which panics on a width mismatch, aborting the worker
  task for a whole batch.

Tests

New coverage: 30 tests across state_override.rs, rpc_batch.rs,
errors.rs, contracts.rs and the eleven new datasets. The regression tests
for the contracts hash transposition, the non-indexed Upgraded log and the
batch-chunking property were each verified by reintroducing the bug and
confirming the test fails.

Every schema block in the new book pages was diffed byte-for-byte against
triodion help <DATASET> rather than written by hand. All eleven match.

The request-count claim was measured through a counting proxy against
mainnet: blocks -b 21000000:+500 sends 6 HTTP requests batched against 501
unbatched, and the two parquet files are byte-identical.

cargo +nightly fmt --all --check          # clean
cargo clippy --workspace --all-targets --all-features -- -D warnings   # exit 0
cargo test --workspace                    # 207 passed, 0 failed
cargo check --manifest-path crates/python/Cargo.toml                   # exit 0

crates/python is workspace-excluded, so --workspace silently skips it; it
is checked separately because this change adds three Args fields that its
adapters construct.