fix(test): regenerate corrupted db fixtures and fix grpc server startup race

[kariy]

The katana-grpc tests in CI have been consistently failing. The root cause turned out to be two separate bugs, one masking the other.

Bug 1: Corrupted database fixtures

The generate_migration_db binary that produces the spawn_and_move.tar.gz and simple.tar.gz fixture archives was creating the database via Db::in_memory() which uses SyncMode::UtterlyNoSync — a mode that does not guarantee committed data is flushed to disk. The binary then archived the database files while the MDBX environment was still open and the node still running, capturing an inconsistent on-disk state. This manifested as:

thread 'test_get_state_update' panicked at crates/utils/src/node.rs:99:63:
failed to open database: failed to open db environment: MDBX_CORRUPTED: Database is corrupted

The fix switches generate_migration_db to use a persistent database directory with SyncMode::Durable, stops the node before archiving, and excludes the non-portable mdbx.lck lock file from archives. The fixture archives have been regenerated on Linux x86_64 inside the ghcr.io/dojoengine/katana-dev:latest Docker container.

Two regression tests (open_spawn_and_move_db_fixture, open_simple_db_fixture) verify the fixtures can be opened without corruption:

# Against OLD fixtures (from main):
        FAIL    katana-utils node::tests::open_simple_db_fixture
        FAIL    katana-utils node::tests::open_spawn_and_move_db_fixture
        fixture database is corrupted: MDBX_CORRUPTED: Database is corrupted

# Against NEW fixtures (regenerated with the fix):
        PASS    katana-utils node::tests::open_simple_db_fixture
        PASS    katana-utils node::tests::open_spawn_and_move_db_fixture

Bug 2: gRPC server startup race condition

Once the database corruption was fixed, the gRPC tests still failed with Connection refused. The gRPC server's start() method was spawning serve_with_shutdown on a background task and returning immediately — before the TCP port was bound. The test's GrpcClient::connect() fired before the server was listening. This was always a latent bug but was masked by the MDBX_CORRUPTED error which prevented the node from starting at all.

The fix binds the TcpListener eagerly before spawning the server task, then passes it via serve_with_incoming_shutdown. This guarantees the server is accepting connections when start() returns and correctly resolves port 0 to the actual assigned port — matching how the RPC server already works.