Add widechar ODBC call tests (driving Unicode fix)

[fdcastel]

Summary

Adds a dedicated test file tests/test_wide_errors.cpp that exercises both ConvertingString constructor scenarios in MainUnicode.cpp:

• Output path — SQLGetDiagRecW, SQLErrorW (uses ConvertingString(length, sqlState) to widen the error text back to the caller's SQLWCHAR buffer).
• Input path — SQLExecDirectW, SQLPrepareW (uses ConvertingString(connection, wcString, length) / convUnicodeToString to narrow the caller's SQLWCHAR query into bytes).

Why this PR exists (and why it's separate from #289)

Per @irodushka's plan in #289 (comment):

1. ASAN/Valgrind CI stays strictly scoped to Add ASAN and Valgrind integration for CI test suite #289.
2. This PR provides the regression coverage for the known Linux widechar bugs (truncation to one SQLWCHAR on read-back, stack smash on small buffers).
3. A follow-up PR will ship the MainUnicode.cpp rewrite and un-skip these tests on Linux.

Linux behaviour (today)

Every test in the new file starts with:

SKIP_ON_LINUX_WIDECHAR();

which is a GTEST_SKIP with a pointer to issue #287 Tier 1b. The pattern matches existing per-platform skips in tests/test_wchar.cpp:111 and tests/test_errors.cpp. This keeps Linux CI green; Windows runs them for real and they must pass. Removing the skip is the visible diff in the follow-up Unicode-fix PR.

Test plan

• CI green on Linux (tests show as SKIPPED)
• CI green on Windows (tests run and pass)
• Locally remove the SKIP_ON_LINUX_WIDECHAR() guards on Linux and confirm the tests fail the way we expect (truncation / overrun) — that's what makes them useful as driving tests for the follow-up PR.

Related

• Issue Adopt ASAN / Valgrind in the test suite #288, PR Add ASAN and Valgrind integration for CI test suite #289 (ASAN/Valgrind CI).
• Issue [EPIC] vNext Migration Plan #287 Tier 1b (Unicode rewrite — consumed by the follow-up PR).

[irodushka]

@fdcastel

Great job.
I'll try to implement an improved version of conv +- within a week.