Docker compose: add resource limits, restart policy, and fix OTLP metrics export

[eldios]

Problem

Docker compose validator deployments have no CPU/memory limits, no restart policy, and a broken OTLP metrics export configuration:

1. No resource limits — containers consume all available resources. On the OVH test validator (8-core/30GB), ScyllaDB alone used 195% CPU and 14GB RAM, causing load averages of 14+ and degraded performance for all services.

2. No restart policy — when shards OOM or crash, they stay down until manually restarted. This causes the validator to become partially unavailable.

3. OTLP metrics export broken — the Alloy config has explicit Content-Type: application/x-protobuf header and compression = "gzip" in the Prometheus OTLP exporter, causing HTTP 400 errors from the remote endpoint. The otelcol.exporter.otlphttp sets these automatically; the explicit values conflict.

Solution

• Add configurable CPU/memory limits via LIMIT_CPUS_* / LIMIT_MEM_* env vars with sane defaults for an 8-core/32GB machine
• Add restart: unless-stopped to all long-running services
• Remove explicit Content-Type header and gzip compression from alloy-config.river
• Increase shard memory default to 2560M (1536M caused OOM kills under normal cross-chain message load)