test(StdJson): cover readBytes at parseRaw's inferred-type lengths

[developer3516]

Motivation

Closes #592.

The reported bug — readBytes returning wrong data or reverting on 20-byte
values — does not reproduce on master. It was real on forge-std v1.2.0, where
readBytes was abi.decode(vm.parseJson(json, key), (bytes)). It now calls
vm.parseJsonBytes directly, which sidesteps the cause.

The cause itself is still there, one layer down. parseRaw infers a value's
type instead of taking a declared one:

hex string length	parseRaw output	inferred as
1, 19, 21, 31, 33 bytes	96 bytes	dynamic bytes
exactly 20 bytes	32 bytes	address
exactly 32 bytes	32 bytes	bytes32

So abi.decode(json.parseRaw(key), (bytes)) still reverts at exactly 20 and 32
bytes. readBytes avoids it, but nothing in the suite says so — readBytes has
no test coverage at all — and nothing in the docs warns about it.

Solution

• test_ReadBytesAtInferredTypeLengths asserts readBytes round-trips 19, 20,
  21, 31, 32 and 33 byte values, bracketing both inference boundaries. If
  readBytes is ever routed back through parseRaw, this fails.
• A @dev note on parseRaw documents the inference, so the next person
  decoding its output as bytes knows why it reverts.

Testing

forge test: 207 passing, 0 failing. forge fmt --check clean. Build matrix
(stable, --via-ir, solc 0.8.13 and 0.8.35, each with and without IR) all green.