Overview

This PR introduces a new telemetry metric machine_category to categorize instantiated compute machines dynamically by identifying their underlying hardware components (e.g., CPU, GPU, TPU). The metric outputs the detected categories similarly to existing telemetry structures maintaining consistency with execution flows. For example, the output traces configurations like a3-highgpu-8g:GPU,tpu7x-standard-4t:TPU,c2-standard-4:CPU.
Here is a section you can append directly to your PR description summarizing the categorization logic clearly for reviewers:

Categorization Logic

The telemetry extraction (detectMachineCategory) identifies the hardware bucket of an instance via a strict, prioritized sequence to handle ambiguous instance shaping (such as -standard- suffixes shared across CPU and GPU families):

1. TPU Inference: Evaluates exactly against the toolkit's native IsTPU() identifier, correctly capturing ct*, tpu, and v* pod slice shorthands directly.
2. GPU Lookup: Resolves natively against config.GetMachineMappings().MachineFamilyToLabelMap. Ambiguous GPU nodes like g2-standard or rtx-6000 (via shorthand substitution) natively map to "nvidia" payload labels and are flagged correctly as GPU.
3. GPU String Keyword: Acts as a safety net explicitly catching standard types directly featuring "gpu" implicitly (like a2-highgpu-1g).
4. CPU Lookup: With accelerators ruled out, we evaluate against all known standard CPU families using config.GetMachineMappings().CpuMachineFamilies (e.g. n1-, c2-).
5. CPU Fallback: Finally, it checks against uniform GCP structural classifiers (-standard-, -highmem-, -highcpu-, custom-). This operates purely as a fallback to guarantee that completely undocumented new CPU families securely classify as CPU rather than Other until mappings update.

Changes

• New Telemetry Metric: Introduced the MACHINE_CATEGORY metric to dynamically categorize compute machines as CPU, GPU, or TPU within telemetry data.
• Classification Logic: Implemented a prioritized detection mechanism in collector_util.go that leverages existing toolkit mappings and naming conventions to identify hardware types.
• Robust Testing: Added comprehensive unit tests in collector_test.go to validate classification accuracy across various machine types, including shorthand mappings and edge cases.