Fetches time-bucketed data from one of the caller's own custom Node-RED
tables. Same request/response contract as GET /timeseries/{table},
with three differences:

• Bucketing/filtering always operates on inserted_at — the one
  column every custom table is guaranteed to have — regardless of any
  other timestamp column the flow author configured.
• No company_id scoping is applied internally: the schema itself
  (named after the caller's own company_id) is the tenant boundary,
  so there's nothing else to filter by. There is no hierarchy
  inheritance — only the exact company that logged the data can query it.
• The requested date range (created_at_gte/created_at_lte, or the
  default 7-day window) cannot exceed 6 months. Custom tables don't
  have the same operational guarantees (chunk sizing, retention,
  indexing) as core platform tables, so an unbounded range is a real
  scan-cost risk.

Not yet supported (tracked as a fast-follow): pivot_on,
pivot_values, pivot_top_n, percentile aggregations (pX, median),
and histogram. Requesting any of these returns 400.

Use GET /timeseries/custom/{table}/schema first to see which columns
exist and whether they're usable as a metric (numeric, aggregatable).