feat: Add ignore_task_fleet_capacity_drifts to task_instance_fleet

[lucasrenan]

Issue

EMR Managed Scaling adjusts Task fleet capacity units as the cluster is running.

Terraform later detects drift and forces them back to the originally declared low target_on_demand_capacity / target_spot_capacity, triggering unintended scale downs during terraform applies.

Root causes

• EMR Managed Scaling mutates live capacity.
• Terraform treats initial targets as authoritative desired state.
• Reconciling causes shrink.
• lifecycle.ignore_changes must be a static list (cannot be conditionally used).

Workaround

Provide two mutually exclusive aws_emr_instance_fleet resources.
One with lifecycle.ignore_changes on the two capacity attributes selected via a boolean variable. When the ignore variant is active, EMR owns scaling and Terraform stops planning destructive reductions.

---

This solution duplicates code and isn't elegant. Any other ideas?