Fix per-thread throughput when EStats are disabled

[Muhammad Labeeb Aryan bin Mohd Lokman (Labeeb2339)]

Fixes #24

Cause

In v5.40, worker start/end timestamps were captured only inside the tcp_row block. tcp_row is created only for successful TCP EStats setup, so normal runs without -es left per-thread worker time at zero even though per-thread byte counters and totals were populated.

Change

• Capture worker start/end timestamps independently of optional EStats collection in both synchronous and asynchronous worker paths.
• Keep EStats calls and availability gated by a valid tcp_row.
• Leave per-thread byte accounting, totals, and throughput measurement formulas unchanged.

Validation

• Added tests/test_per_thread_throughput.ps1, a dependency-free regression guard for both worker paths, and wired it into the Windows build workflow.
• Regression checks pass locally.
• Built Release x64 locally with the available Visual Studio toolchain (using UndockedSourceLink=false, UndockedUseDriverToolset=false, and SpectreMitigation=false because those local components are unavailable).
• Ran rebuilt v5.40 sender/receiver loopback tests without -es in synchronous and asynchronous modes; both reported non-zero per-thread time/throughput and non-zero totals.

Related upstream work exists in #31; this change keeps the fix scoped to the issue and preserves existing EStats behavior.