feat: migrate Langfuse filter pipeline to SDK v4

[hassiebp]

Summary

• replace the Langfuse v3 filter example with a Python SDK v4 implementation
• preserve one logical trace per OpenWebUI chat with deterministic trace IDs
• propagate user, session, tags, and trace metadata before creating observations
• record parented user-input spans and model generations with model, input, output, and v4 usage_details
• close turn roots on outlet, replacement, valve updates, and shutdown, then flush pending data

The filter declares langfuse~=4.7, which pip resolves as >=4.7,<5. Pipelines splits frontmatter requirements on commas, so this compatible-release spelling preserves the upper bound without being parsed as two packages.

Linear: LFE-14968

Related Langfuse docs PR: langfuse/langfuse-docs#3503

Validation

• Ruff formatting: 1 file already formatted
• python3 -m py_compile examples/filters/langfuse_v4_filter_pipeline.py — passed
• Langfuse Python SDK v4.14.2 in-memory OpenTelemetry smoke test:
  • Requirement range passed: langfuse~=4.7 accepts 4.x and rejects 5.0
  • OTel continuity/propagation passed: 3 spans, 1 trace, 2 direct children

Scope

This does not address unrelated historical OpenWebUI tokenization, model-prefix, or tool-calling behavior.