bug: SQLITE_TOOBIG crash when running wrangler dev --local with multiple zones

## Environment

- Running mode: `wrangler dev --local` (Kubernetes sidecar, not deployed to Cloudflare Workers)
- Zones: 10
- wrangler version: bundled in Docker image

## Problem

When running the exporter via `wrangler dev --local`, the `colo-metrics` and `load-balancer-metrics` Durable Objects crash on every alarm cycle with:

```
{"logger":"metric_exporter","namespace":"account::colo-metrics","level":"error","msg":"Refresh failed","error":"string or blob too big: SQLITE_TOOBIG"}
{"logger":"metric_exporter","namespace":"account::load-balancer-metrics","level":"error","msg":"Refresh failed","error":"string or blob too big: SQLITE_TOOBIG"}
```

This repeats every ~60 seconds indefinitely, eventually causing wrangler to crash with:

```
✘ [ERROR] Uncaught Error: Network connection lost.
```

## Root cause

`MetricExporter` stores its entire state as a **single blob** via `ctx.storage.put(STATE_KEY, this.state)`. For `colo-metrics`, the state contains `metrics: MetricDefinition[]` with data for all zones × all PoPs, which grows into a multi-MB JSON blob. miniflare's local SQLite storage has a blob size limit that this exceeds.

`load-balancer-metrics` hits the same issue because it stores per-zone load balancer pool weight data for all zones.

## Impact

- These two DOs never successfully persist state
- Their alarm cycles flood logs with errors
- The accumulated error load eventually causes `Network connection lost` and wrangler crashes
- Pod restarts continuously if not wrapped in a restart loop

## Workarounds

None available without code changes. `METRICS_DENYLIST` only filters Prometheus output — it does not prevent DO creation or data fetching (see related issue).

## Suggested fix

See PR #23 which adds query-name support to `METRICS_DENYLIST`, allowing users to skip DO creation for specific queries entirely. A more complete fix would chunk the stored state across multiple KV keys to handle large datasets.

## Steps to reproduce

1. Deploy with `wrangler dev --local`
2. Configure 5+ Cloudflare zones
3. Observe `colo-metrics` and `load-balancer-metrics` DOs crashing every ~60s

Start with MetricExporter and the ctx.storage.put(STATE_KEY, this.state) call used by the colo-metrics and load-balancer-metrics Durable Objects. Reproduce with wrangler dev --local and 5+ zones, then trace how each object builds its state. Done means alarm cycles persist large datasets without SQLITE_TOOBIG or causing wrangler to lose its network connection.

Written by the indexing model from the issue text.