Use crate::test_rng::Rng in every randomized test: three more xorshift64* copies sit beside the module that claims to be the only one #1868

[agent] Filed by the scheduled architecture audit (audit-core). Register: register.

Kind: refactor (test-only). Register ID: A29.

Problem

test_rng.rs calls itself "the one deterministic RNG the randomized tests share". There are three more copies of the same xorshift64* generator:

• crawlers/oracle_support.rs L10-L38:`` pub(crate) struct Rng, with the same `new`/`next`/`below`/`pick` and a `chance(pct: usize)` spelled `below(100) < pct`. `test_rng` spells it `next() % 100 < percent` with a `u64` argument.
• patch/copy_tree.rs L237-L251: a private struct Rng(u64) with next and below.
• crawlers/npm_crawler/oracle.rs L1056-L1068:``` next/`below` inlined into a fixture builder.

All four produce the same stream for the same state (the same shifts and multiplier), so a single type serves every caller.

Proposed change

• Delete the three copies. Use crate::test_rng::Rng in the crawler oracle tests, copy_tree's randomized test and the npm oracle builder.
• Keep test_rng's chance(u64) and change the crawler callers' argument type.
• Seeds stay the same, so every generated case is unchanged.

Size

4 files, about −50 lines, all #[cfg(test)].

Acceptance criteria

• No production change.
• Every randomized test in crawlers/* (the *_matches_the_async_oracle tests), patch::copy_tree and npm_crawler passes with the same seeds and the same case counts.
• grep -rn "0x2545_F491_4F6C_DD1D" crates/socket-patch-core/src returns one hit, in test_rng.rs.

Not in scope (noted in the register)

The eight crawlers/**/oracle.rs files, 2,978 lines with oracle_support.rs, keep the pre-blocking-pool crawlers as equivalence oracles. cargo_crawler/oracle.rs already differs from production: it reads with tokio::fs::read_to_string, which follows FIFOs and links, while production uses read_regular_to_string_sync. Whether to delete these oracles is a separate question.