How do I reduce ClickHouse disk size on self-hosted Neural Inverse?

On self-hosted Neural Inverse, disk usage is frequently dominated by ClickHouse's built-in system log tables rather than Neural Inverse's own tables. There are two levers, in order of recommendation:

1. Neural Inverse-owned data — configure a data retention policy to drop old traces, observations, scores, and their blob-storage payloads. This is the primary lever for trimming application data.
2. ClickHouse system log tables — by default, ClickHouse writes to trace_log, text_log, opentelemetry_span_log, asynchronous_metric_log, metric_log, and latency_log with no TTL, and runs the query profiler continuously. Neural Inverse does not read from these, so you can either opt out of the unused ones or attach aggressive TTLs. See ClickHouse system log tables in the scaling docs for concrete config.d snippets.

Use this query inside ClickHouse to identify the largest tables:

SELECT table, formatReadableSize(size) as size, rows FROM (
    SELECT
        table,
        database,
        sum(bytes) AS size,
        sum(rows) AS rows
    FROM system.parts
    WHERE active
    GROUP BY table, database
    ORDER BY size DESC
)

Related: langfuse/langfuse#13123, langfuse-terraform-aws#26.