Add capability-based gating for ODBC e2e tests (mssql-odbc)

Problem statement

The ODBC e2e DriverConnectLiveTest fixture gates every live test on ODBCTestConfig::HasConnection(), which is true whenever any connection method is configured (ODBC_TEST_CONNSTR, ODBC_TEST_DSN, or ODBC_TEST_SERVER). Most tests are fine with "any connection string that authenticates."

But some tests have stronger requirements. The connection-string parser-parity tests (MalformedTokenReturnsSuccessWithInfo, ConnectionStringParserParityBehaviors) assemble their own login strings from cfg.Server()/Uid()/Pwd() and deliberately corrupt individual auth tokens. In environments configured via ODBC_TEST_CONNSTR, a DSN, or integrated auth, Uid()/Pwd() are legitimately empty, so these tests would build broken empty-credential strings and fail even though the suite is configured correctly.

A tactical guard was added in PR #107 (ODBCTestConfig::HasSqlAuth() + GTEST_SKIP) to stop the spurious failures. This issue tracks the proper design: a small capability model so tests can declare what they need instead of hand-checking individual fields.

Proposed solution

Introduce a capability model on ODBCTestConfig and let each test declare its requirements:

• enum class ODBCCap { SqlAuth, MutableDatabase, IntegratedAuth, Dsn, Encrypt };
• bool ODBCTestConfig::HasCapability(ODBCCap) const;
  • SqlAuth = Server && Uid && Pwd all present
  • other values wired as best-effort predicates for future tests
• A one-line per-test skip helper (macro-backed, because GTEST_SKIP() must return from the test body to actually skip):
  • ODBC_REQUIRE_CAP(ODBCCap::SqlAuth); at the top of a test
• Base-connection reuse so auth-neutral tests append to the configured connection instead of rebuilding it:
  • std::string ODBCTestUtils::BuildConnectionStringNarrow() delegating to the existing BuildConnectionString() (single source of truth), plus the existing SqlTString overload.

Then split the parser-parity tests:

• Auth-neutral cases (assert 01S00, connection still succeeds) append malformed tokens to BuildConnectionStringNarrow() and run against any working config — no capability gate.
• Identity-corruption cases (assert SQL_ERROR + 28000, and first-wins duplicates) declare ODBC_REQUIRE_CAP(ODBCCap::SqlAuth).

Note on first-wins: keep those UID-based under SqlAuth rather than converting to duplicate Database keys — BuildConnectionString() already emits a Database= key and the parser is first-wins, so an appended Database=bad;Database=good would lose to the base's existing Database and silently break the negative case.

Affected crate

Not sure / Multiple

Alternatives considered

• Keep the tactical HasSqlAuth() + inline GTEST_SKIP (current state in PR #107): works, but every future special-requirement test re-implements its own field checks; doesn't scale.
• A dedicated SqlAuthLiveTest fixture whose SetUp() skips unless SQL auth is present: reasonable, but a capability enum is more flexible when a test needs a combination of capabilities and keeps the requirement visible at the call site.

Additional context

Tactical fix commit: f457cff on saurabh500-odbc-connstr-parser (PR #107). The parity tests live in mssql-odbc/tests/e2e/tests/driver_connect_test.cpp; the fixture/config are in mssql-odbc/tests/e2e/include/odbc_test_fixture.h and lib/odbc_test_config.cpp; the base builder is ODBCTestUtils::BuildConnectionString() in lib/odbc_test_utils.cpp.

Read mssql-odbc/tests/e2e/tests/driver_connect_test.cpp, mssql-odbc/tests/e2e/include/odbc_test_fixture.h, lib/odbc_test_config.cpp, and lib/odbc_test_utils.cpp, starting with the existing HasSqlAuth guard and connection-string builder. Add the capability checks and reuse the base connection string, then split the parser-parity cases so auth-neutral tests run with any configured connection and identity-corruption cases require SqlAuth.