test(gateway): register with public keys the gateway can accept

[kvinwang]

Problem

Ten of the twenty-eight cases in gateway/test-run/test_suite.sh cannot pass. Every one of
them registers a CVM with a hand-written placeholder public key, and every one of those keys
is the wrong length:

"error": "[cafebabe] invalid client public key: public key is 41 characters, expected 44"

A WireGuard public key is 32 bytes, so 44 base64 characters. The thirteen placeholders in the
suite are 38 to 43:

Placeholder	Length
bootnode12345678901234567890123456789=	38
threenode12345678901234567890123456789=	38
stresstest%02d12345678901234567890123456=	39
partition%02d123456789012345678901234567=	39
backup123456789012345678901234567890123=	40
persist%02d123456789012345678901234567890=	40
upnode…, uptest…	40
prpctest…, waltest%02d…, downtest…, downnode…	41
persisttest1234567890123456789012345678901=	43

They were fine until registration started validating the key length
(WG_PUBLIC_KEY_B64_LEN in gateway/src/kv/import.rs, added with the recovery-API review
follow-ups). The suite is not in CI — CI runs test_proxy.sh — so nothing said so.

The cases this silences are the ones that cover persistence and replication most directly:
test_wal_integrity, test_periodic_persistence, test_client_registration_persistence,
test_stress_writes, test_network_partition, test_three_node_cluster,
test_three_node_bootnode, test_prpc_register, and both node-status registration cases.
test_periodic_persistence never reaches its persistence assertion at all.

Fix

Each placeholder becomes a call to test_public_key, the generator the suite already has and
that the other five call sites already use. It emits a real 32-byte key, so the gateway
accepts it, and it is deterministic, so a failure is still reproducible. Seeds are distinct
per registration — two instances sharing a public key are refused as a conflict, which would
just be the same class of bug again.

The generator's doc comment now says why a placeholder is not a shortcut, so the next one
does not get written.

Verification

Before, on next: 18 passed, 10 failed — the ten above.

After, each previously-failing case run individually:

test_prpc_register                       PASSED
test_client_registration_persistence     PASSED
test_stress_writes                       PASSED
test_network_partition                   PASSED
test_wal_integrity                       PASSED   (Registered 5/5 clients, WAL 944 bytes)

with the rest still running at the time of writing; I will update this if any of them turn
out to be failing for a second, unrelated reason.

shellcheck -S error clean, bash -n clean. No production code is touched.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[kvinwang]

All ten now pass, run individually against a release build of next + this change:

test_prpc_register                       PASSED
test_client_registration_persistence     PASSED
test_stress_writes                       PASSED
test_network_partition                   PASSED
test_three_node_cluster                  PASSED
test_three_node_bootnode                 PASSED
test_periodic_persistence                PASSED
test_node_status_register_exclude        PASSED
test_node_status_register_reject         PASSED
test_wal_integrity                       PASSED   (Registered 5/5, WAL 944 bytes)

None of them turned out to have a second, unrelated reason to fail — the key length was the
only thing standing in front of all ten.