UPSTREAM PR #1186: enable flash attn by default

[loci-dev]

Mirrored from leejet/stable-diffusion.cpp#1186

[loci-review]

Explore the complete analysis inside the Version Insights

I've generated a comprehensive summary report for your project. Here are the key highlights:

Critical Findings:

⚠️ Performance Regressions Detected - The analysis shows significant performance degradation across 10 key functions:

Most Critical Issues:

1. std::vector<ggml_backend_reg_entry>::end() - 226.52% response time increase
2. std::_Rb_tree::_S_key() - 164.20% response time increase
3. std::vector::_S_max_size() - 108.86% response time increase
4. std::swap() [nlohmann::json] - 76.37% response time increase

Root Causes:

• 70% of issues are in STL container operations
• Potential compiler optimization or ABI compatibility issues
• Memory allocation overhead concerns

Recommendation:

❌ NOT RECOMMENDED FOR MERGE - Pull Request #15 requires immediate investigation and remediation of these performance regressions before proceeding.

The report includes detailed metrics, analysis, and specific recommendations for addressing these issues.