feat(parser): support multiple tables in a single markdown document

- Add extract_table_blocks() to split a document into contiguous table
  blocks, ignoring prose, headings, and blank lines between them
- Add parse_document() as the new top-level entry point that runs
  extract_table_blocks + detect_has_duration_column + parse_table per
  block and returns a combined flat list of rows
- Guard against empty End cells (e.g. in-progress rows) by validating
  the end field before calculating duration
- Update cli.py to use parse_document() instead of the manual
  detect + parse combo
- Add tests for extract_table_blocks and parse_document, including two
  smoke tests against the real 2026-W21 weekly timesheet file

src/timesheets/cli.py

@@ -4,7 +4,7 @@ import sys
 from datetime import date
 
 from .output import print_summary, write_csv
-from .parser import aggregate_rows, detect_has_duration_column, parse_table
+from .parser import aggregate_rows, parse_document
 from .projects import load_project_map
 from .utils import format_date
 
@@ -18,7 +18,8 @@ def build_parser() -> argparse.ArgumentParser:
         help="Path to the markdown file containing the timesheet table, or '-' to read from stdin.",
     )
     parser.add_argument(
-        "-o", "--output",
+        "-o",
+        "--output",
         help="Path to the output CSV file. Defaults to stdout.",
         default=None,
     )
@@ -59,7 +60,7 @@ def main() -> None:
             sys.exit(1)
 
     lines = content.splitlines()
-    rows = parse_table(lines, has_duration_col=detect_has_duration_column(lines))
+    rows = parse_document(lines)
 
     if not rows:
         print("Warning: no timesheet rows found in input.", file=sys.stderr)