Fix DiskSpd parser for v2.2 intermediate CPU-topology headers (exactly-64-vCPU case)

[AlexWFMS]

Summary

Follow-up to #733. DiskSpd 2.2's CPU-utilization table prefixes the CPU column with a dynamic, hierarchical set of topology columns — Socket, Node, Group, Core, Class — and each one is emitted only when the system has more than one of that unit (see DiskSpd ResultParser.cpp::_PrintCpuUtilization):

[Socket |] [Node |] [Group |] [Core |] [Class |] CPU |  Usage |  User  | Kernel |  Idle

The previous fix special-cased only two exact header strings — the full Socket | Node | Group | Core | CPU and the bare Core | CPU. Any intermediate combination fell through to the Contains("Core | CPU") branch, which inserted the CPU section title mid-line, so the table sectionized under a key like Socket | Node | CPU and parsing crashed:

Results parsing failed for 'diskspd' workload. The given key 'CPU' was not present in the dictionary.

This is the customer-reported "works on <64 and >64 vCPU, fails on exactly 64" case: a 64‑vCPU VM with multiple NUMA nodes but a single processor group emits an intermediate header with no Group column.

Fix

Replace the hard-coded header prefixes with NormalizeCpuTable, which:

1. Locates the CPU table by its invariant trailing signature CPU | Usage | User | Kernel | Idle (independent of leading columns).
2. Collapses whatever leading topology columns are present down to the canonical [Group |] CPU | Usage | ... form.
3. Titles the section CPU so it sectionizes correctly.

The Group column is retained (when present, i.e. > 64 vCPUs) so ParseCPUResult still maps the group-relative CPU number to a unique processor id; all other leading columns (Socket/Node/Core/Class) are dropped. Keying off the trailing signature also future-proofs against new columns such as Class on heterogeneous-core (P/E) systems. Downstream ParseCPUResult/ProcessAndUpdateString are unchanged.

Validation on real Azure VMs (range of vCPU sizes)

Captured authentic DiskSpd 2.2 output across a range of VM sizes in West US 2:

VM size	vCPU	socket / node / group	CPU header	Pre-#733	This PR
D8as_v5	8	1 / 1 / 1	Core | CPU	✅	✅
D16as_v5	16	1 / 1 / 1	Core | CPU	✅	✅
D32ds_v5	32	1 / 1 / 1	Core | CPU	✅	✅
D64as_v5	64	1 / 1 / 1	Core | CPU	✅	✅
E64ds_v5	64	2 / 2 / 1	Socket | Node | Core | CPU	❌ crash	✅
D96as_v5	96	1 / 2 / 2	Node | Group | Core | CPU	❌ crash	✅
E96ds_v5	96	2 / 2 / 2	Socket | Node | Group | Core | CPU	✅	✅

The two bold intermediate headers are exactly the combinations the old code crashed on. Both are now committed as authentic test fixtures.

Tests

• Added DiskSpdParserVerifyV220FormatOnMultiSocketSingleProcessorGroupSystem (authentic Socket | Node | Core | CPU from E64ds_v5 — the "exactly 64 vCPU" case).
• Added DiskSpdParserVerifyV220FormatOnMultiNumaMultiProcessorGroupSystem (authentic Node | Group | Core | CPU from D96as_v5, with a group boundary verifying group‑1 CPUs get unique ids).
• Both new fixtures were confirmed to fail on the pre-fix parser with the exact customer error, and pass with this change.
• All 40 DiskSpd tests pass (dotnet test --filter FullyQualifiedName~DiskSpd).

VERSION bumped 3.3.12 → 3.3.13.