perf: allow concurrent dataset pushes by narrowing the push-path locks

[vdusek]

Concurrent pushes to a dataset were serialized end to end, so N concurrent pushes cost N API round-trips instead of overlapping. Two locks were wrapping the network I/O, and because either one alone is enough to serialize, both had to be narrowed for the win to materialize:

• Actor.push_data took charging_manager.charge_lock() unconditionally and held it across await dataset.push_data(...).
• ApifyDatasetClient.push_data additionally held its per-client _lock across the chunked push_items calls.

The charge lock exists to keep the limit reservation and the charge that acts on it atomic, which can only matter for pay-per-event runs. ApifyDatasetClient.push_data mutates no client state at all - _check_and_serialize is a classmethod, _chunk_by_size only uses locals, and push_items is a stateless API call - so outside of charging there is nothing for either lock to protect.

Both call sites now go through a single charge_lock_if_charging() helper in _charging.py, keyed off charging_manager_ctx. That replaces the near-duplicate DatasetClientPpeMixin._charge_lock and the second, divergent is_pay_per_event predicate, so the "do we need the lock?" decision lives in one place for all three call sites. Pay-per-event runs are unchanged and still fully serialized.

Two behavior notes:

• ApifyDatasetClient pushes and drop() are no longer mutually exclusive. The lock only ever ordered them within a single client instance in a single process, so a concurrent drop was never actually safe; _lock still serializes drop() itself.
• Local runs on the file-system client stay serialized by Crawlee's FileSystemDatasetClient._lock, which guards sequential item numbering and metadata. That lock is load-bearing and there is no network round-trip behind it.

Regression tests at both layers, each verified to fail on master: test_concurrent_push_data_overlaps puts a barrier inside the mocked push_items and proves the API calls now overlap, and test_push_data_does_not_take_charge_lock_without_pay_per_event pins the Actor.push_data half. test_concurrent_actor_push_data_stays_within_budget guards the opposite direction - it fails if the charge lock is ever skipped for pay-per-event runs.

✍️ Drafted by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.06%. Comparing base (cad8d7e) to head (ef63509).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1079      +/-   ##
==========================================
+ Coverage   91.83%   92.06%   +0.22%     
==========================================
  Files          51       51              
  Lines        3235     3237       +2     
==========================================
+ Hits         2971     2980       +9     
+ Misses        264      257       -7     

Flag	Coverage Δ
e2e	35.31% <37.50%> (+<0.01%)	⬆️
integration	57.08% <75.00%> (+0.05%)	⬆️
unit	84.18% <100.00%> (+0.96%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.