fix(content): generalize joy-check writing examples and de-ai the rubric

[notque]

Summary

• Replaces 6 overly-specific examples (all about architecture provenance from one blog post) with diverse topics that each teach a different grievance pattern
• Runs de-ai pass on structural prose and JOY examples (em-dashes, hedging, adverb crutches)

New example topics

#	Pattern	Topic
1	Victim vs. learner	Project failure
2	Accusation vs. curiosity	Duplicate work discovery
3	Demanding reciprocity vs. quiet satisfaction	Open-source thanklessness
4	Entitlement vs. self-knowledge	Passed over for recognition
5	Escape vs. forward motion	Career transition
6	Bad-faith attribution vs. sitting with discomfort	Ambiguous feedback

De-AI fixes

• 9 em-dashes removed from structural prose and JOY examples
• "honestly" replaced with "directly", "I think" rephrased to "I understand"
• 3 remaining em-dashes are intentional in GRIEVANCE (FAIL) examples

Test plan

• scan-ai-patterns.py shows 3 remaining hits, all in GRIEVANCE examples
• All 6 rubric dimensions covered across examples
• 3/3 review rounds passed