Some test suites, like sharding_jscore_passthrough_with_balancer, enable random chunk migrations, which behind the scenes enables a fail point that triggers migrations without the usual requirements, such as large enough chunks.

However, when this fail point is enabled, the balancer appears to disregard the sharding zones defined for collection.

For example, the test case attached to this ticket creates two zones, each linked to a shard, and waits for the chunks to be placed according to the zones. This will quickly converge on the sharding_jscore_passthrough suite (balancer enabled, no random migrations) but it will often diverge on the sharding_jscore_passthrough_with_balancer (balancer enabled + random migrations) suite with the test ultimately timing out.

To provide better coverage with zones on passthrough suites, the random migrations should respect the defined zones, i.e. prioritize splits on the zone boundaries and not migrate across zones.