fix(tests): stabilise e2e suite on macOS CI runners

[grumbach]

Context

The E2E suite has been recurrently failing on macos-latest with:

test_attack_corrupted_public_key panicked:
quotes should succeed: InsufficientPeers(
  "Got 6 quotes, need 7. Failures: [<peer>: timeout: Timeout waiting for quote from <peer>]"
)

The same root cause surfaces as test_chunk_exists, test_chunk_put_duplicate_skips_payment, test_payment_required_enforcement and a few others. Always 6-of-7 on macOS; Linux runners pass consistently.

Why only macOS

This looked weird — "it's slow networking" doesn't explain a platform split. It turns out to be test-infrastructure, not a client bug:

1. Zero redundancy in the testnet. CLOSE_GROUP_SIZE is 7, and DEFAULT_NODE_COUNT was CLOSE_GROUP_SIZE + 1 = 8. Quote collection gates on quotes.len() >= CLOSE_GROUP_SIZE, i.e. all 7 of the remote peers (the client is the 8th) must answer. One slow peer = fail. The 2x over-query is a no-op because only 8 nodes exist.

2. macOS GitHub runners are roughly half the CPU throughput of Linux runners. They're nested-virt on Anka/VMware. Spawning 8 QUIC nodes in-process and handshaking to all of them simultaneously saturates the available CPU during the handshake burst. On Linux, all 8 handshakes finish comfortably under the 10s per-peer timeout. On macOS, one or two routinely don't — the test sees 6 successes, one timeout failure, and no way to reach quorum.

Put together: the test had no slack at all and was running on the one platform that couldn't meet the timing. Merkle was unaffected because it already bumped its own config to 120s.

Fix

Two changes, both test-only — prod ClientConfig::default() is unchanged:

1. DEFAULT_NODE_COUNT from CLOSE_GROUP_SIZE + 1 (8) to CLOSE_GROUP_SIZE * 2 (14). One full extra group of slack. Up to 7 peers can be slow and the test still reaches quorum. Extra cost: ~1.2 s of spawn delay per test setup.

2. New test_client_config() helper in tests/support/mod.rs with quote_timeout_secs = store_timeout_secs = 60. All e2e_* files except e2e_merkle.rs (which already had its own 120 s config) switch from ClientConfig::default() to this helper. 60 s is deliberately conservative — in the happy path everything completes in ~1 s, the extra budget only shows up on flakes.

Both constants carry doc comments explaining why they're different from production, since the numbers look surprising at a glance.

Test plan

• cargo fmt --all --check: clean
• cargo clippy --all-targets --all-features -- -D warnings: clean
• cargo test -p ant-core --test e2e_chunk test_chunk_put_get_round_trip: passes locally (macOS)
• cargo test -p ant-core --test e2e_security test_attack_corrupted_public_key: passes locally — this is the one that was flaking
• cargo test -p ant-core --test e2e_payment test_payment_required_enforcement: passes locally — the other flake
• Full CI matrix will run on this PR; expect macOS E2E to finally go green consistently