[Feature] Source-aware subjective feedback ingestion

[shchukins]

Parent

Part of #82.

Idea

Support subjective feedback from multiple sources while preserving origin and scale semantics.

Sources

• Telegram inline buttons
• future iOS app
• Strava perceived exertion if available
• manual input

Problem

Different sources may use different scales and meanings. The system needs normalized values while preserving original payloads.

Proposed solution

For each feedback record store:

• feedback_source
• normalized feedback_score
• feedback_value
• original_source_value
• original_source_payload
• mapping version

Strava scope

If Strava perceived exertion is available:

• import it
• map to internal normalized scale
• preserve original value
• avoid overwriting explicit Telegram feedback unless source precedence is defined

Acceptance criteria

• source metadata is persisted
• normalized mapping is documented
• source precedence rules are explicit
• tests cover mapping behavior