UPSTREAM PR #17969: webui: Improve copy to clipboard with text attachments

[loci-dev]

Mirrored from ggml-org/llama.cpp#17969

Close #17834

• Implemented functionality that allows to copy & paste user message including the pasted text
• Added a Setting that disables separation of prompt text + attached text and copies/pastes all as a single text string
• Rearranged copy utils into copy.ts file & supplimented with unit tests
• Reorganized unit tests location in webui project

Demo

demo.mp4

[loci-review]

Explore the complete analysis inside the Version Insights

Performance Analysis Summary - PR #537

Analysis Result

This PR introduces WebUI-only changes for clipboard handling of chat messages with text attachments. All modifications are in TypeScript/Svelte frontend code within tools/server/webui/. Zero C/C++ code changes were made to llama.cpp core libraries, GGML tensor operations, or inference pipelines.

Performance Impact: The binary analysis confirms zero change in power consumption across all llama.cpp binaries (libggml-base.so, libggml-cpu.so, libllama.so, llama-run, etc.). No functions within Performance-Critical Areas (matrix operations, attention mechanisms, model loading, tokenization, sampling) were modified. The changes are isolated to browser-side clipboard utilities that execute only on user copy/paste events.

Inference Impact: No impact on tokens per second. Functions responsible for tokenization and inference (llama_decode, llama_encode, llama_tokenize, ggml_mul_mat, llama_graph_compute) remain unchanged with identical response time and throughput metrics between versions.