Dictionaries are replicated per NUMA node, so memory usage scales with the number of NUMA nodes spanned by the instance. The thresholds used in the test were written with the assumption that there's a single copy of each dict.
Fix that.
Fixes SCYLLADB-2320
Should be backported to releases since 2025.2, in case someone wants to run the CI of those on NUMA machines.
Closesscylladb/scylladb#30178
* github.com:scylladb/scylladb:
test_sstable_compression_dictionaries: fix the test when running on more than one NUMA node
test/pylib: add get_shard_to_numa_node_mapping()
api: add system/shard_to_numa_node_mapping
2576ce1f83