Fix stream_blocks dropping column info for empty result sets

[jenish-25]

Fixes #222.

Problem

For a SELECT over an empty table, QueryResult::stream_blocks() yields no blocks at all, so callers lose all column/schema information. fetch_all() is unaffected.

Root cause

ClickHouse precedes a query's data blocks with a single schema-only header block (columns present, num_rows == 0). stream_blocks() deliberately skips this first block (via block_index > 1) because for non-empty results the following data blocks already carry the full schema, so the empty header would just be noise.

When the result set is empty, however, the header is the only block the server sends before EOF. Skipping it leaves the stream empty and the caller has no way to learn the columns/types. (fetch_all() avoids this because it drives the internal stream with skip_first_block = false.)

Fix

Buffer the header block instead of discarding it:

• If a real data block arrives later, the buffered header is dropped — behaviour is unchanged for non-empty results (only data blocks are streamed).
• If EOF is reached without any data block, the buffered header is emitted so its column info is preserved — empty results now stream exactly one block with the columns and zero rows.

Testing

Added test_empty_select_stream_blocks, which streams SELECT 1 AS a WHERE 1 <> 1 and asserts a single 1-column, 0-row block is produced. Confirmed it fails before this change and passes after.

Verified against a local clickhouse-server; the new test and the existing streaming/empty-select tests pass, with no regressions introduced by this change.