test(body.history): de-flake idempotency assertion with fake timers

The `should clone body before transformation to preserve original` test
asserted `expect(result1).toBe(result2)` for two consecutive
transformRequestBody calls on identical input. That invariant is not
guaranteed by production code because the cch billing hash in
src/headers/billing.ts:59 mixes `Date.now().toString(36)` into its input
to mimic CC's per-request attestation hash. Two calls in different
milliseconds produce different cch values and therefore different output
bytes.

The test only ever passed when both calls landed in the same millisecond —
reliable on bare-metal fast hardware, unreliable under husky pre-commit +
pre-push with lint-staged overhead, CI workers, or any system under load.
It happened to be stable enough to not trip on previous 0.1.4 work, then
shifted across the threshold during the 0.1.4 release sequencing and
blocked the push.

Fix: wrap the two calls in `vi.useFakeTimers()` + `vi.setSystemTime(...)`
so the frozen clock produces byte-identical cch values and the
idempotency assertion becomes meaningful (it validates that NOTHING
other than time-derived fields varies across calls, which is the actual
invariant the test name promises). Real timers are restored in finally
so the rest of the suite is unaffected.

Verification: body.history.test.ts 5/5 green in isolation, full vitest
suite 3/3 green (933 tests each run).