Fix node memory usage accuracy: use memoryUsage instead of workloadMe…

[si1enteee]

…moryUsage

Problem:
The node list view uses workloadMemoryUsage (container_memory_working_set_bytes) as the primary memory metric, which includes reclaimable Page Cache. This causes:

1. Inflated memory usage for nodes with large Page Cache (e.g. Elasticsearch, database nodes showing 72% when actual pressure is 48%)
2. Usage exceeding 100% when working_set > allocatable
3. Jump discontinuity when switching between Prometheus and Metrics Server data sources (working_set vs actual usage differ by 24%+)
4. Inconsistency with the detail view which already shows both metrics separately

Additionally, the memoryUsage PromQL uses MemTotal - MemFree - Buffers - Cached which incorrectly treats tmpfs/shared memory in Cached as reclaimable, causing underestimation for nodes with large tmpfs or shared memory segments.

Fix:

1. Switch renderMemoryUsage primary metric from workloadMemoryUsage to memoryUsage (actual node memory usage excluding reclaimable cache)
2. Add workloadMemoryUsage as a tertiary line (LineProgress tertiaryValue) so working set / OOM risk information is preserved
3. Add workload info to tooltip for context
4. Update memoryUsage PromQL from: MemTotal - (MemFree + Buffers + Cached) to MemTotal - MemAvailable which correctly accounts for non-reclaimable cache (tmpfs, shm)

The detail view (node-charts.tsx) already correctly shows both metrics separately; this change brings the list view in line with that behavior.

Verified with simulated data for 6 scenarios:

• ES data node: 72.4% to 48.3% (fixes inflation)
• tmpfs node: 54.3% to 79.3% (fixes underestimation)
• Shared memory node: 48.8% to 85.7% (fixes underestimation)
• High load node: 108.5% to 95.0% (fixes >100%)
• Normal/low load nodes: minimal change