[XRT-SMI] Update TCT tests to follow different logic from strix and report throughpuy

[aktondak]

Problem solved by the commit

The TCT validate tests did not handle MDS (NPU3) correctly: they assumed fixed Strix token counts when computing latency and throughput. This change adds an MDS path that uses runner elapsed time and iteration count, keeps Strix behavior unchanged, and shares a common workload runner between both platforms. Separately, xrt-smi watch mode now refreshes the terminal on each step for snapshot-style reports (examine, context-health), while firmware-log and event-trace streaming watch modes are unchanged. The shim also maps firmware log configure failures to a clear “Invalid firmware log level” message.

Bug / issue (if any) fixed, which PR introduced the bug, how it was discovered

None

How problem was solved, alternative solutions (if any) and why they were rejected

TCT tests use one run_tct_workload() path and branch only in calculate_tct_metrics() (Strix uses fixed samples × iterations; MDS uses iterations only). We rejected duplicating full Strix/MDS run functions and using the runner’s precomputed latency/throughput fields, which do not match the MDS workload model. Watch mode defaults refresh_terminal to true, with firmware-log and event-trace explicitly opting out.

Risks (if any) associated the changes in the commit

The reported TCT numbers would now change

What has been tested and how, request additional testing if necessary

Tested on npu3 platform. This TCT throughput seems close to no-op throughput so numbers make more sense now :

C:\Users\aktondak>xrt-smi validate --advanced -r tct-all-col
-------------------------------------------------------------------------
                    DISCLAIMER  (xrt-smi --advanced)
You are running a developer command that may change system configuration.
                Continue only if you understand the risks.
-------------------------------------------------------------------------
Validate Device           : [00c5:00:01.1]
    Platform              : NPU Medusa
    Power Mode            : performance
-------------------------------------------------------------------------------
Test 1 [00c5:00:01.1]     : tct-all-col
    Details               : Average TCT throughput (all columns): 10273.5 TCT/s
    Test Status           : [PASSED]

Documentation impact (if any)

None

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"