feat: add linguistic analysis tab for prompting style and patterns

[eeshansrivastava89]

Overview

Add a new dashboard tab that analyzes how users write their prompts — the language, patterns, and style. The current dashboard covers volume, timing, and persona. This would be the "level 2" that answers: what does your prompting actually look like?

Should be available for all sources combined or filtered per chat source.

Potential visualizations (pick 3–4 for v1)

High confidence — likely compelling

• Top phrases / n-grams: Most-used 2–3 word phrases (e.g. "can you", "make sure", "let's try"), ranked by count with frequency bars. This is the kind of thing people screenshot and share.
• Prompt openers: How you start prompts — categorize into buckets (direct instruction, question, context-setting, greeting) and show distribution as a donut or bar chart.
• Directive vs. interrogative ratio: Are you a "do X" person or a "what if we…" person? Simple ratio with a visual scale.

Medium confidence — interesting but may need iteration

• Vocabulary richness: Type-token ratio, average words per prompt, reading level. Simple stats that are fun to compare across sources.
• Prompt length distribution: Histogram of prompt lengths. Are you a one-liner or a paragraph writer?
• Per-source comparison: Side-by-side view showing how your language differs across Claude vs. Cursor vs. Copilot. This could be the most interesting view — people write differently to different tools.

Design considerations

• Keep it to 3–4 visualizations max in v1 — each one should be immediately "oh that's interesting" without needing explanation
• Match the existing coffee-themed dashboard aesthetic
• The per-source breakdown is key for shareability — people will want to see how they talk to different tools
• This is pure text analysis (tokenization, counting) — no embeddings needed, so it should be fast

Depends on

• feat: validate persona axis robustness and recalibrate distributions #24 (persona axis validation) — not a hard dependency, but the linguistic tab should complement the persona system rather than overlap
• feat: consolidate Human in the Loop and Vibe Coder into a single flagship metric #25 (metric consolidation) — should be aware of the final metric set to avoid redundancy