nanodbc: handle SQL_NO_TOTAL when retrieving binary data in chunks

[jeroen]

Fixes #1024.

Problem

When retrieving unbound SQL_C_BINARY columns, nanodbc reads the value in 1024-byte chunks via repeated SQLGetData calls. The loop in get_ref_impl<std::vector<std::uint8_t>> only appended a chunk when the length indicator was positive:

if (ValueLenOrInd > 0) { ... out.insert(...) ... }

Per the ODBC spec, a driver may instead set the indicator to SQL_NO_TOTAL (−4) on intermediate SQL_SUCCESS_WITH_INFO calls when it does not know the remaining data length — which is what the Databricks/Simba Spark driver does. Since −4 fails the > 0 test, every full 1024-byte chunk was silently discarded, and only the final chunk (returned with SQL_SUCCESS and a real length) was kept. That yields exactly the behavior reported in #1024: n mod 1024 bytes returned, with exact multiples of 1024 surviving intact.

(One note on the issue's analysis: the returned bytes are the last chunk, not a prefix — the reporter's prefix check used a period-256 payload with a length that is a multiple of 1024, where the two are byte-identical.)

Fix

Handle SQL_NO_TOTAL by appending the full buffer, exactly mirroring the existing SQL_C_CHAR and SQL_C_WCHAR blob paths a few lines above (which already have this branch). On SQL_SUCCESS_WITH_INFO/01004 with SQL_C_BINARY the driver fills the entire buffer, so appending buffer_size is correct.

Testing

There is no regression test because reproducing this requires a driver that reports SQL_NO_TOTAL for binary chunks (the Simba Spark driver does; the drivers in CI report actual remaining lengths, which take the already-correct ValueLenOrInd > 0 path). Verification against a Databricks warehouse per the repro in #1024 would be welcome.

Upstream nanodbc main has the identical gap in the binary path; I'd suggest upstreaming this patch as well.

🤖 Generated with Claude Code

[jeroen]

Upstreamed to nanodbc as nanodbc/nanodbc#475.

[andrie]

Removing the comment. I meant to post this in #1024

[simonpcouch]

Yup, looks good. Thank you!