tests: fix NLB replacement test bootstrap crash due to missing rackdc properties

[mykaul]

Summary

• Fix TestFullNodeReplacementThroughNlb::test_should_survive_full_node_replacement_through_nlb which always crashes when bootstrapping node 4/5
• Root cause: _bootstrap_node() calls ccm_cluster.add() without data_center or rack — CCM leaves cassandra-rackdc.properties empty, and Scylla's snitch crashes with locator::bad_property_file_error
• Fix: infer data_center/rack from existing cluster nodes and pass explicitly

Details

The test was added in PR #706 with @skip_scylla_version_lt(scylla_version="2026.1.0"). Since CI uses SCYLLA_VERSION: release:2025.2, the test was always skipped and the bug was never caught.

Scylla error from node4 startup log:

ERROR snitch_logger - Failed to parse a properties file (cassandra-rackdc.properties). Halting...
ERROR init - Startup failed: locator::bad_property_file_error (std::exception)

Testing

Verified 10/10 passes on Scylla 2026.1 after the fix (previously 0/10).

[dkropachev]

@mykaul , this fix is wrong - test needs to specify concrete dc and rack it wants to place new nodes in, taking them from nodes already provisioned

[mykaul]

@mykaul , this fix is wrong - test needs to specify concrete dc and rack it wants to place new nodes in, taking them from nodes already provisioned

@dkropachev - isn't it correct anyway, to specify the topology explicitly? I can fix the test in addition, but this seems correct as well.

[dkropachev]

@mykaul , this fix is wrong - test needs to specify concrete dc and rack it wants to place new nodes in, taking them from nodes already provisioned

@dkropachev - isn't it correct anyway, to specify the topology explicitly? I can fix the test in addition, but this seems correct as well.

The problem is that test will dump all new nodes into the same dc+rack, which is wrong, it should spread them through existing racks.

[mykaul]

v2: Addressed review feedback from @dkropachev.

Instead of inferring data_center/rack from the first existing node (which is semantically wrong for multi-DC/multi-rack topologies), _bootstrap_node now accepts explicit data_center and rack parameters. The caller passes the known topology value (data_center='dc1') directly, making the intent clear at the call site.

Changes:

• _bootstrap_node(self, ccm_cluster, node_id) → _bootstrap_node(self, ccm_cluster, node_id, data_center=None, rack=None)
• Removed inference logic (existing = next(iter(...)) / or 'dc1' / or 'RAC1')
• Call site now passes data_center='dc1' explicitly
• Rebased on current master