[MeABzZz]

What remains after main

Josh's a694465 ("cap RPC bodies…") already moved the HTTP/1 JSON-RPC read onto http_body_util::Limited, so an unauthenticated client can no longer make the node buffer an unbounded body. This PR is now rebased onto that change with a merge commit, and only the remaining delta is left:

• main caps the read at the compile-time DEFAULT_MAX_REQUEST_SIZE (1 MiB), so a configured max_request_size above 1 MiB can never be used. Any larger body gets a 413 before the configured check runs.
• this PR changes one line, Limited::new(req.into_body(), server.max_request_size), so the server stops reading at the configured cap, whether that is smaller or larger than 1 MiB.

Diff vs main: src/rpc/server.rs, the one-line cap change plus two tests.

Verification (Rust 1.88.0, branch head 8641580, includes node main 4f1be96)

test	main's line (mutation)	this branch
http1_accepts_body_under_a_cap_above_the_default (2 MiB cap, ~1.5 MiB body)	FAILED: 413 Payload Too Large	ok, 200
http1_rejects_oversized_body_before_the_remainder_is_sent	FAILED: Elapsed (waits for the rest of the body)	ok

• Also passing on the branch: lib rpc:: 161, rpc_tests 29, rpc_edge_cases 9, rpc_advanced_scenarios 9.
• Full cargo test --no-fail-fast: 3016 passed / 2 failed / 26 ignored (main 4f1be96: 3016 / 0 / 26). Both failures are in modules this PR doesn't touch (storage::ibd_engine::disk_index::…async_spill_serves_queries_via_pending_then_registers, node::parallel_ibd::chunk_assigner::…a6m_gd_slow_open_uncools_prior_hero_when_pin_empty). Rerun on this branch, they passed 5 out of 5 times, so they are timing-dependent under the full parallel run. The other three reconciled branches had 0 failures.
• cargo clippy --all-targets: the same 90 diagnostics as main, with 0 new. Main already fails lib-test clippy on two nonminimal_bool errors in src/node/parallel_ibd/chunk_assigner_tests.rs (5044, 5105), which this PR doesn't touch.
• rustfmt --check on src/rpc/server.rs: clean. Note that main itself isn't cargo fmt --check clean (tests/feature_flags_reflection_tests.rs:7, already there on 4f1be96).

Review notes

• Builds on: a694465 (Limited read). Merged with main at 4f1be96; no force-push.
• Merge order: independent of the other open node PRs (git merge-tree is clean against fix(node): anchor assume-valid per network with Core's hash and minimum work #31, fix(node): count witness bytes in the mempool cap at insert #34, fix(node): reject mempool txs over standard weight #35, fix(network): bound inbound frames and drop bad checksums #36, fix(network): ban public peers for oversized messages #37, fix(node): refuse a reorg over empty undo data and tell the operator at startup #38 and Optional: pass assume-valid ancestry to consensus (companion to blvm-consensus#24) #39).
• Decision: none. If 1 MiB is meant as a hard ceiling whatever the config says, close this PR instead.