perf(metrics): batch ClickHouse inserts in /ingest/batch endpoint

[EmirUfak]

Problem

In fluxer_metrics/src/api/ingest.rs, ingest_batch inserts each metric
one by one. For ClickHouse, that means one INSERT per metric, even though
the client already sent them together in a single batch.

So a payload with 100 metrics causes 100 inserts instead of 1.

Proposed solution

Add batch methods to the Storage trait:

• insert_counters_batch
• insert_gauges_batch
• insert_histograms_batch

For ClickHouseStorage, open one client.insert(...) per metric type,
write all rows, then call end() once. Other backends can keep the
current behavior via a default trait impl.

Update ingest_batch to call the new methods.

Expected impact

• Inserts per request: from N down to at most 3.
• Latency:lower p50/p95 on /ingest/batch, especially for bigger batches.
• Throughput: higher, since we stop paying per item round-trip cost.
• ClickHouse load: fewer, larger inserts = fewer parts + less merge
  pressure.

Small batches (1–2 items) won't see a meaningful change.

Notes (optional)

• No API or schema changes.
• On failure, a whole metric-type batch is rejected instead of per-item.
  Seems fine since the client already groups them, but worth flagging.

Happy to open a draft PR if this sounds good.

Checks

• I searched for existing discussions and didn't find a duplicate.