test(settlement): cover the settle_all MAX_SETTLE_ALL boundary

[thlpkee20-wq]

Closes #299

`record_usage` caps the per-agent service index at
`MAX_AGENT_SERVICE_INDEX` (== `MAX_SETTLE_ALL`), so `settle_all`'s
`SettleAllTooLarge` guard had zero test coverage — it can't be
reached through the public API alone (its own doc comment says as much:
it exists to protect a future migration that could write a larger
index).

Added two tests that write an oversized index directly via
`as_contract`, mirroring the existing `..._via_storage_helper` pattern
used for other invariants in this file:

• exactly-at-boundary (256 services): succeeds, settles all of them.
• one-over-boundary (257 services): panics with `SettleAllTooLarge`
  (Add tests for lifetime usage counters and last-settlement timestamps #19).

A real finding along the way

The exactly-at-boundary test needed `env.budget().reset_unlimited()` —
settling 256 services in one call (2 storage reads + 2 writes + 1 event
each) exceeded the default test budget around service #178. That's a
test-harness-only accommodation for this PR, but it's a genuine signal:
a `settle_all` call anywhere near the configured `MAX_SETTLE_ALL` would
likely also exceed real network resource limits on-chain. Flagging this
as worth a follow-up look at whether `MAX_SETTLE_ALL` should be
tightened — that's a behavior change I did not make here, out of scope
for a test-coverage PR.

Test plan

• `cargo fmt --all -- --check`
• `cargo clippy --all-targets -- -D warnings`
• `cargo test` — 337 passed, 1 failed. The one failure is the same
  pre-existing, unrelated test covered by fix(test): correct cross-call event-count assumption in ownership-transfer test #315.

[mikewheeleer]

neat work @thlpkee20-wq — thanks for the care. merging 🌱