fix: governance gaps surfaced by encode implementation bugbot review

[klappy]

Four governance gaps caused four code bugs in klappy/oddkit#96. The server had no governance to follow for these cases and improvised — bugbot caught the improvisations.

Fixes

1. Removed bare "found" from observation trigger words — collided with Learning's "found that" during fallback regex classification
2. Added fallback: true to observation.md frontmatter — observation is the canonical fallback for unmatched paragraphs
3. Added ## Fallback Behavior section to how-to-write-encoding-types.md — specifies how fallback type is resolved via frontmatter
4. Added ## Scoring Algorithm section — centrally defines score→level mapping (strong=max, adequate=60%+, weak=40%+, insufficient<40%)
5. Added ## Context vs Input section — input generates artifacts, context only informs quality scoring

Prompt over code requires complete governance. Gaps force the server to improvise.

---

Note

Medium Risk
Although this is doc-only, these governance changes can alter runtime encode classification and quality scoring wherever the server dynamically follows these articles.

Overview
Adds explicit governance for encode edge-cases by documenting fallback type resolution, score-to-quality level thresholds, and how context is used only for quality scoring (not artifact creation).

Marks odd/encoding-types/observation.md with fallback: true and removes the bare trigger word found to reduce misclassification during unstructured fallback parsing.

^{Reviewed by Cursor Bugbot for commit 53ef9c7. Bugbot is set up for automated code reviews on this repo. Configure here.}