[1.0] Minimize and redact trace data before export

Trace serialization currently carries strings and dictionary values into exports without a general capture-time limit or redaction boundary. Protecting the dashboard alone cannot protect data already written to traces.

Scope

• Define and apply a bounded diagnostic representation before trace/log export: retain useful operational fields and exclude raw protected payloads by default.
• Redact known credentials throughout nested supported values, reuse the lm_kwargs fix, and prevent diagnostic serialization errors from falling back to raw values.
• Keep explicit trusted diagnostic access separate from generally exported observability data.

Acceptance criteria

• Synthetic secrets and private payload canaries are absent from exported spans/log records, including nested containers and exception paths.
• Record-size limits are observable; useful error codes, run identifiers and timing remain available.
• Tests cover capture/export output itself rather than only a filtered UI response.

Boundaries

No blanket claim that arbitrary strings can always be recognized as secrets; unapproved raw content must be omitted rather than relying only on pattern matching.

References

• Parent topic: #280.
• Implementation dependencies: #383, #428.
• src/flock/logging/trace_and_logged.py
• src/flock/engines/dspy_engine.py
• src/flock/api/collector.py

Start by reading src/flock/logging/trace_and_logged.py, then trace the related paths in src/flock/engines/dspy_engine.py and src/flock/api/collector.py. Done means exported spans and log records omit synthetic secrets and private payloads, including nested and exception data, while retaining useful diagnostics and exposing record-size limits; tests must verify capture/export output.