fix(tests): environment-agnostic cassette normalization and sorting

[tychtjan]

Problem

When VCR cassettes are recorded locally (Docker Compose) vs on staging, the resulting YAML files differ in ways that have nothing to do with actual API changes — causing noisy, hard-to-review PRs.

Root causes identified:

1. The API returns certain arrays (referenceProperties, edges, workspaceDataFilterReferences, etc.) in non-deterministic order
2. XLIFF/XML localization groups come back in different order
3. Export hash placeholders get session-global numeric indices that shift with test execution order
4. The demo user has profile fields (email, name) locally but not on staging

Changes

Targeted JSON array sorting

Sorts only arrays under 5 specific keys known to be non-deterministic: referenceProperties, edges, workspaceDataFilterColumns, workspaceDataFilterReferences, userGroups. Applied only during YAML serialization — does not affect VCR replay behavior or test assertions.

XML group sorting fix

_sort_xml_groups() was only called for str response bodies. Compressed responses arrive as bytes and were skipped entirely. Now applied in both branches of custom_before_request() and custom_before_response().

Export hash stabilization

Replaces session-global numeric indices (EXECUTION_NORMALIZED_3) with format-based names (EXECUTION_NORMALIZED_CSV, EXPORT_NORMALIZED_XLSX). Each cassette gets self-contained, deterministic placeholders regardless of test execution order.

Demo user profile fields

Added firstname, lastname, email to demo_user.json so make load-staging produces the same user profile as the local Docker bootstrap.

Test plan

• Recorded all cassettes locally with OVERWRITE — all tests pass
• Replayed without OVERWRITE — all tests pass (190 SDK + 66 pandas)
• Verified make clean-staging + make load-staging succeeds and demo user has profile fields
• Verified targeted sorting only touches safe keys; user/workspace/data-source ordering is preserved

risk: low

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.18%. Comparing base (19e1cc1) to head (fcd47aa).

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1521   +/-   ##
=======================================
  Coverage   78.18%   78.18%           
=======================================
  Files         228      228           
  Lines       14930    14930           
=======================================
  Hits        11673    11673           
  Misses       3257     3257           

☔ View full report in Codecov by Sentry.
📢 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.