add metrics + per-runtime layer variants

[mmosarafO2]

• Metrics: OTLP receiver now handles /v1/metrics alongside /v1/traces; new src/enhanced_metrics.rs synthesizes aws.lambda.enhanced.{duration, billed_duration, init_duration, restore_duration, memory_size, max_memory_used, memory_utilization, invocations, errors, timeouts, out_of_memory, cold_starts} from Telemetry API platform.report events. Datadog Lambda Enhanced Metrics parity, all runtimes, no code change.
• Optional O2_EMIT_BASE_ALIASES=true also emits aws.lambda.{duration, errors, invocations} so Datadog-shaped dashboards render without a separate CloudWatch Metric Streams ingest.
• gzip outbound bodies (logs + OTLP) via new src/compress.rs.
• Preserve OTLP Content-Type end-to-end (protobuf pass-through).
• otel-instrument: runtime-aware; detects nodejs / python / java from AWS_EXECUTION_ENV and wires the corresponding SDK loader. Lambda-tuned defaults: OTEL_METRICS_EXPORTER=otlp, OTLP protocol http/protobuf, OTEL_METRIC_EXPORT_INTERVAL=1000, OTEL_BSP_SCHEDULE_DELAY=500.
• build.sh: per-runtime (node/python/java/core) x per-arch (x86_64/arm64) variants. Backward-compat alias o2-lambda-extension-.zip -> node bundle. New `./build.sh repackage` skips the Rust rebuild.
• deploy.sh rewritten: publishes openobserve-extension--
  per variant with correct compatible-runtimes; DEPLOY_LEGACY_ALIAS=1 keeps publishing the pre-variant name for backward compatibility.
• Bugfixes: log-flush errors no longer block span/metric flush in either sync or async path; request_id fallback to record.requestId for platform.report; async flush wrapper timeout raised 500ms -> 3s; HTTP client timeout now honors O2_REQUEST_TIMEOUT_MS everywhere.
• README overhauled: signals table, per-runtime variants, enhanced metrics reference, ESM caveat, Datadog coverage comparison, testing tiers.