ref(api): Rework how replicator resources are done

[iambriccardo]

Summary

• Rework ETL API Kubernetes resource defaults so k8s.replicator_resources uses the clearer DefaultReplicatorResourcesConfig type and requires only baseline replicator CPU/memory requests.
• Keep pipeline replicator_resources fully optional with ReplicatorResourcesConfig: pipelines can override CPU/memory requests and/or explicit CPU/memory limits, while omitted limits are derived from the final request values using the static ETL API multipliers.
• Compute full Kubernetes resource settings through ReplicatorStatefulSetResourcesConfig, using API defaults plus optional pipeline overrides for the replicator container and static constants for the vector container; final requests are clamped to the API's minimum emitted CPU/memory values and limits are never below their paired requests.
• Require source connection settings, including TLS settings, to be present in ETL API configuration instead of relying on implicit source defaults.
• Validate default replicator resource requests immediately after the API config is loaded during ETL API startup, keeping application construction free of config validation plumbing.
• Document the default-vs-pipeline override behavior in the ETL API README and align skip_serializing_if = "Option::is_none" usage across related ETL API config/response types.
• Simplify BigQuery BIGNUMERIC scale validation by reading the parsed numeric scale directly instead of formatting the value as a string.

Validation

• cargo x fmt
• cargo check -p etl-api --all-targets --all-features
• cargo nextest run -p etl-api --all-features -E 'test(replicator_stateful_set_resources) | test(pipeline_replicator_resources_are_persisted_and_used_on_start) | test(updating_a_running_pipeline_reapplies_replicator_resources) | test(updating_a_stopped_pipeline_only_persists_replicator_resources) | test(invalid_replicator_resources_are_rejected) | test(stored_pipeline_config_serialization) | test(stored_pipeline_config_deserializes_without_replicator_resources)'
• cargo nextest run -p etl-api --all-features -E 'test(test_create_ducklake_replicator_stateful_set_json) | test(test_create_bq_replicator_stateful_set_json) | test(test_create_iceberg_replicator_stateful_set_json) | test(test_create_init_containers)'
• cargo clippy -p etl-api --all-targets --all-features -- -D warnings

[coveralls]

coverage: 72.775% (+0.05%) from 72.721% — improve-bigquery into main