fix(dataset): keep CSV export columns aligned

[anxkhn]

export_csv_to_stream (src/crawlee/_utils/file.py) built the CSV header from the
first non-empty item's keys, but then wrote every row positionally with
csv.writer and item.values(). Dataset.push_data accepts an arbitrary
Mapping[str, JsonSerializable] per item with no schema enforcement, so a dataset
where later items have a different key set or key order than the first (normal when
different handlers push different fields) had its values written under the wrong
columns, or shifted into columns that do not exist, silently corrupting the CSV.

Example: pushing

{'name': 'Alice', 'age': 30}
{'name': 'Bob', 'city': 'NYC', 'age': 25}
{'age': 40, 'name': 'Carol'}

and exporting with content_type='csv' produced

name,age
Alice,30
Bob,NYC,25
40,Carol

Bob's row has three values in two columns, and Carol's row puts age (40) under
name and name (Carol) under age.

This change writes rows with csv.DictWriter, keyed on the first item's fields and
extrasaction='ignore', so each value lands in its own column and keys not in the
header are dropped. The same input now exports as

name,age
Alice,30
Bob,25
Carol,40

which matches the column-aligned behavior of Crawlee for JavaScript (whose CSV
export, by default, derives the header from the first item, maps each row by key,
and ignores extra keys). Items missing a header key get an empty cell, as before for
absent values.

The existing lineterminator='\n' default (which avoids double line endings on
Windows) is preserved: it still flows through **kwargs into DictWriter.

Issues

• No existing issue; this is a self-reported bug found while reading the dataset
  export path. (Happy to open a tracking issue first if the maintainers prefer.)

Testing

• Added two regression tests in tests/unit/_utils/test_file.py:
  • test_export_csv_to_stream_keeps_columns_aligned_for_heterogeneous_items -
    asserts values stay under their own header column for items with differing key
    sets/orders (fails on the old positional code with the exact misalignment above).
  • test_export_csv_to_stream_skips_empty_items - empty mappings are skipped and do
    not define or shift the header.
• uv run pytest tests/unit/_utils/test_file.py tests/unit/storages/test_dataset.py
  plus the test_basic_crawler.py push-and-export tests: 273 passed, 2 skipped
  (unrelated memory-storage persistence skips).
• uv run poe lint (ruff format + check) and uv run poe type-check (ty) are clean
  for the changed files.

Checklist

• CI passed

[Mantisus]

Thank you for finding this bug. However, we cannot use extrasaction=‘ignore’

[Copilot]

Pull request overview

This PR fixes CSV export for Dataset items that don’t share identical key order/sets, preventing values from being written under the wrong column when exporting heterogeneous items.

Changes:

• Update export_csv_to_stream to write rows by key (via csv.DictWriter) instead of positional item.values().
• Add regression tests covering heterogeneous item keys/order and skipping empty items during header selection.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/crawlee/_utils/file.py	Switches CSV export implementation to key-based writing and changes how CSV headers are derived.
tests/unit/_utils/test_file.py	Adds regression tests for column alignment with heterogeneous items and for skipping empty mappings.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[anxkhn]

pushed a revision addressing the latest review and ci feedback.

[vdusek]

Same as here #2074 (comment)

@anxkhn Thank you. For future reviews, I'd really appreciate it if you could address all the feedback in a single follow-up commit, rather than splitting the changes across multiple commits, merging them into earlier commits, and force-pushing it. Your current approach makes the changes really hard to review.

[vdusek]

A few important things and then a few nits.

[anxkhn]

@vdusek addressed all new threads, including js-default parity, streaming behavior, bounded default memory, and value preservation tests. rebased and re-requesting review. thank you.