fix: prefix ClusterId with project_id to prevent collisions across projects (GTI-685)

[ymendez-redis]

Problem

GTI-685

When memorystore.py output from multiple GCP projects is combined, redis2re.py merges instances that share the same ClusterId and Region into one row — losing data and producing incorrect memory calculations.

This happens because Valkey and Redis Cluster instances use a short name as ClusterId (e.g., memorystore-valkey), which can be identical across projects. Redis standalone is unaffected because GCP returns a full resource path (projects/proj-id/locations/.../instances/name).

Real production example

Project	ClusterId	Region	Result
project-019	cluster-011	europe-west2	Unique — different region
project-010	cluster-011	us-central1	Merged with project-004 ❌
project-006	cluster-011	europe-west1	Unique — different region
project-004	cluster-011	us-central1	Merged with project-010 ❌

10 input rows → 3 output rows instead of 4.

Root Cause

GCP returns different identifier formats per instance type:

Instance Type	GCP Label	Format	Collision risk
Redis standalone	instance_id	projects/proj-id/locations/.../instances/name	✅ Globally unique
Valkey	instance_id	memorystore-valkey (short name)	❌ Collides across projects
Redis Cluster	cluster_id	memorystore-redis-cluster (short name)	❌ Collides across projects

Fix

Extracted a _resolve_inst_key() helper that detects short vs full GCP paths:

• Full GCP paths (starting with projects/) → used as-is (already unique)
• Short names → prefixed with project_id/ (e.g., my-project/memorystore-valkey)

Updated _accumulate_commands, _attach_memory_usage, and _attach_capacity_scalar to use the helper.

Why fix in memorystore.py and not redis2re.py

• redis2re.py processes multiple integrations (ElastiCache, OSStats). Changing its grouping logic risks breaking them.
• The problem originates in memorystore.py — it's the only source producing non-unique ClusterIds.
• ElastiCache and OSStats are unaffected (their IDs are already unique).

Verified end-to-end pipeline

Stage	Impact
memorystore.py output	ClusterId becomes project-id/cluster-name for Valkey/Cluster
Conversion to xlsx	Passes through as-is
redis2re.py grouping	metric_key = (clusterId, region) — now unique across projects
redis2re.py get_db_name	Strips / via regex — Salesforce name still valid
msstats.py	Not affected — only processes Redis standalone

Testing

• 10 new unit tests covering: Redis full path, Valkey/Cluster short name prefixing, cross-project collision prevention, empty project_id, unknown fallback, _attach_memory_usage and _attach_capacity_scalar integration
• Verified against live GCP instances in redislabs-sales-pivotal and redislabs-university
• All 23 tests pass, black formatting clean