Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/flashbots.zig`:
- Around line 564-576: The JSON parsing functions currently use
std.heap.page_allocator directly (e.g., checkRpcError), which breaks the
project's allocator threading and harms test/production memory tracking; change
these functions (checkRpcError, parseBundleHashResult, parseCallBundleResult) to
accept an allocator parameter (e.g., allocator: *std.mem.Allocator) and use that
allocator when calling std.json.parseFromSlice, then update all callers in Relay
to pass self.allocator so allocations are tracked consistently.

---

Nitpick comments:
In `@src/flashbots.zig`:
- Around line 108-133: The Bundle struct currently stores slices in txs without
copying, so callers must keep backing memory alive; update the code by adding a
clear doc comment on Bundle (and on addTransaction/transactions) that explains
the ownership model and lifetime requirement of []const u8 entries, and
optionally implement a copying alternative (e.g., addTransactionCopy that
allocates via allocator, copies signed_tx into owned storage and appends that
slice) so users who expect ownership can use it; reference symbols: Bundle, txs,
addTransaction, transactions and the allocator used for any copying routine.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/flashbots.zig`:
- Around line 36-45: Change the CallBundleOpts.state_block_number from ?u64 to
an optional string type so callers can pass hex numbers or tags like "latest":
update the field signature in CallBundleOpts (currently state_block_number:
?u64) to state_block_number: ?[]const u8 (or to a small discriminated union type
e.g., BlockTag = union(enum){Number: u64, Tag: []const u8} and then
state_block_number: ?BlockTag) and then adjust any code that serializes or reads
state_block_number (serialization/JSON building for the Flashbots API) to accept
and forward either the string tag or a hex-encoded number appropriately. Ensure
default null behavior remains unchanged.
- Around line 631-634: The parsing function parseHexOrIntU256 only handles hex
strings and must accept base-10 decimal strings used by Flashbots for fields
like bundleGasPrice/coinbaseDiff/ethSentToCoinbase/gasFees; modify
parseHexOrIntU256 so its string branch checks if the string starts with "0x" and
parses as hex otherwise parses as a decimal integer (using the uint256 parsing
utility or equivalent), and ensure the callers (variables bundle_gas_price,
coinbase_diff, eth_sent, gas_fees in src/flashbots.zig) continue to use this
function; also update the test case to supply decimal-format strings (not
0x-prefixed hex) to match the documented Flashbots response.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/flashbots.zig`:
- Around line 389-392: The code is appending caller-provided strings
(replacement_uuid, privacy.hints, privacy.builders) directly into the JSON
buffer via buf.appendSlice/append which can break JSON or allow injection;
change the code that handles opts.replacement_uuid in the function that builds
the RPC body and the blocks that append privacy.hints and privacy.builders to
first JSON-escape the string (properly escape quotes, backslashes, and control
chars) and then append the escaped result instead of the raw input, reusing or
adding a helper like jsonEscape(allocator, src) and replace calls around
buf.appendSlice/append accordingly, and add a regression test that supplies
values with quotes/backslashes/control characters to verify the output is valid
JSON and fields are preserved.
- Around line 660-669: The parseDecimalU256 function can miss some overflows
because checking result < prev after result = result *% 10 +% digit is
insufficient; instead, before doing the multiply/add, check overflow by
comparing result against the u256 max thresholds: compute max_div_10 = u256.max
/ 10 and max_rem = u256.max % 10, then if result > max_div_10 or (result ==
max_div_10 and digit > max_rem) return error.InvalidResponse; only then perform
result = result *% 10 +% digit. This replaces the post-wrap check and uses the
u256 overflow bounds to prevent truncated/wrapped values in parseDecimalU256.