Fixing a bug related to RAM limits for process_centr_segment step

[sergii-mamedov]

1. Change max_memory=8196 to max_memory=4096. First there was the bug (8196 -> 8192), which caused this step to run with 2048, 4096 and 8192 MB of RAM. The latter value was too large for normal datasets. In the case of datasets with a lot of noise, this volume would still not be enough. But this amount of memory cost us a lot. Second, I limited the maximum value to 4096 MB, as was the original version by Lachlan.
2. I looked at the statistics in the perf_profile_entry table for datasets that used 4096 MB of RAM or more. In most cases, the actual RAM consumption was up to 1.5 GB. This is the most expensive step for us, so changing the base value of runtime_memory like this will save 30-40% of money for heavy datasets. You can view the raw data with the following request:

   SELECT
   id,
   profile_id,
   start,
   finish,
   extra_data ->> 'cost' AS costs,
   extra_data ->> 'runtime_memory' AS runtime_memory,
   extra_data ->> 'max_memory' AS max_memory,
   extra_data ->> 'max_time' AS max_time,
   extra_data
   FROM perf_profile_entry
   WHERE 
   start >= '2024-11-01' AND 
   finish <= '2024-11-30' AND 
   name='process_centr_segment' AND 
   (extra_data ->> 'runtime_memory')::numeric >= 4096
   ORDER BY id DESC;

[sergii-mamedov]

@lmacielvieira some problem with test-graphql