[QNN EP] Address PR #602 review comments (N-1 vacuous tests, T-1 line drift) - #719

Summary

Follow-up to PR #602 review comments from @huaychou.

[N-1] Group 10 ValidateCompiledModelCompatibilityInfo tests were vacuous on x86_64

ValidateCompiledModelCompatibilityInfo in qnn_execution_provider.cc has:

#if !defined(__aarch64__) && !defined(_M_ARM64) && !defined(_M_ARM64EC)
  ORT_CXX_LOG(logger_, ORT_LOGGING_LEVEL_WARNING, "Skip compatibility validation on x86 platforms.");
  *model_compatibility = OrtCompiledModelCompatibility_EP_NOT_APPLICABLE;
  return nullptr;
#endif

Any non-empty info string trips that guard on x86_64, so the previous TooFewFields_NotApplicable and BadVersionFormat_NotApplicable tests were both hitting the same x86-skip branch (empirically verified via ExpectLogged(WARNING, "Skip compatibility validation on x86 platforms.")).

Fix:

• Consolidated TooFewFields_NotApplicable + BadVersionFormat_NotApplicable -> single NonEmptyOnX86Host_LogsSkipAndReturnsNotApplicable.
• Renamed EmptyString_NotApplicable -> EmptyString_LogsNoInfoAndReturnsNotApplicable.
• Both tests now ExpectLogged the actual WARNING each branch emits.
• Group 10 header now documents that the field-parsing branches downstream of the x86 guard are unreachable on this coverage host; extracting the parser to a platform-guard-free free function is a follow-up option if we want unit-tier parser coverage.

[T-1] Absolute line-number references in comments

Replaced absolute source line numbers (line 301 / 547 / 1084) in 5 test comments with function names, so they don't drift when the source file changes. Line 547 had already drifted to 523 and 1084 to 1240 in the current source, illustrating the issue.

Bonus: enable_htp_fp16_clamp_overflow warning branch

Reviewer also asked whether enable_htp_fp16_clamp_overflow needs a unit test. The option was introduced in #686 (merged 2 days before #602) but wasn't picked up by #602's ctor option tests. Added Ctor_HtpFp16ClampOverflowTrueOnUnsupportedSdk_LogsWarning covering the ctor warning branch at qnn_execution_provider.cc:922-926. The coverage build (QAIRT 2.48 = QNN API 2.37) does not define QNN_HTP_FP16_CLAMP_OVERFLOW_AVAILABLE, so the #ifndef branch is compiled in and reachable.

Integration smoke coverage already exists in conv_test.cc:1195 (added by #686).

Test plan

• onnxruntime_provider_test --gtest_filter="QnnUnit_ExecutionProviderTest.*:QnnUnit_ExecutionProviderHtpTest.*" — 65/65 pass
• lint_and_fix clean

Ref: #602, #686