docs: Add LSP UX heuristic evaluation report for clangd

[sinelaw]

Summary

This PR adds a comprehensive usability heuristic evaluation and technical audit of Fresh's LSP integration, focusing on the LSP Status Indicator, Notification Bar, and Hover Pop-ups using clangd as the test language server against the fmt C++ codebase.

Key Changes

• New documentation: docs/internal/LSP_HEURISTIC_EVAL_CLANGD.md — a detailed 160-line evaluation report including:
  • Methodology section documenting the test environment (Fresh v0.2.23, clangd 18.1.3, fmt codebase, tmux 3.4)
  • 15 heuristic violations mapped to Nielsen/Norman usability heuristics with severity ratings (0–4)
  • Technical audit with ANSI color palette analysis, contrast measurements (WCAG AA compliance), and layout integrity checks
  • Remediation plan with 13 actionable recommendations prioritized by impact (P0–P3)
  • Severity roll-up and reproduction instructions

Notable Findings

The evaluation identifies 7 major issues blocking the "LSP should just work" promise:

• LSP does not auto-start; no visual indication of available-but-dormant servers (H-1)
• No progress indicator during clangd indexing (~8 s with no UI feedback) (H-2)
• Dual, inconsistent diagnostic counters in status bar vs. editor (H-3)
• Hover does not surface diagnostic text; requires separate panel navigation (H-5)
• Hover under-informative on qualified names and templates (H-6)
• Hover silently fails on files outside compile_commands.json (H-7)
• Dirty buffer after --no-restore on never-edited files (H-12)

The report also documents 6 minor issues (H-4, H-8–H-11, H-13) and 1 cosmetic issue (H-14), plus 1 strength (acceptable latency under load).

Purpose

This document serves as a baseline for prioritizing LSP UX improvements and provides reproducible test steps for future regression testing.

https://claude.ai/code/session_01E63E6Ye3h5fiRbfGWFcH1F