Update EOS culling to proper regEx substitution

[Gnurro]

Due to more flexible culling by re.sub introduced in July, some model outputs were culled wrongly - this should solve this issue. As the regEx culling is more future-proof, it is kept in the HF backend and added to the llama-cpp backend. Model entries updated accordingly. Since the regEx handling changes how the eos_to_cull key needs to be written, specially important for custom models like finetunes, the documentation on it is updated.

Changes:

• Change EOS culling in llamacpp_api.py to use regEx substitution
• Add model_registry_eos_check.py script for checking model registry entry EOS for regEx
• Update hf/llamacpp model entries' eos_to_cull to expected regEx (escaping |): falcon-7b-instruct, falcon-40b-instruct, oasst-sft-4-pythia-12b-epoch-3.5, openchat_3.5, Yi-34B-Chat, SUS-Chat-34B, openchat-3.5-0106, openchat-3.5-1210, Nous-Hermes-2-Mixtral-8x7B-DPO, Smaug-72B-v0.1, Smaug-34B-v0.1, Qwen1.5-7B-Chat, Qwen1.5-72B-Chat, Phi-3-mini-128k-instruct, Starling-LM-7B-beta, Qwen2-7B-Instruct, Qwen2-72B-Instruct, Llama-3-SauerkrautLM-70b-Instruct, aya-23-8B, aya-23-35B, Meta-Llama-3.1-8B-Instruct, Meta-Llama-3.1-70B-Instruct, Qwen1.5-0.5B-Chat-GGUF-q8, CapybaraHermes-2.5-Mistral-7B-GGUF-q4, CapybaraHermes-2.5-Mistral-7B-GGUF-q5, CapybaraHermes-2.5-Mistral-7B-GGUF-q5-k-s, openchat_3.5-GGUF-q5, Meta-Llama-3-70B-Instruct-GGUF-q4, Meta-Llama-3-70B-Instruct-GGUF-q8, c4ai-command-r-plus-GGUF-q4, c4ai-command-r-plus-GGUF-q8
• Update model registry documentation on eos_to_cull regEx

[Gnurro]

Compared v1.6 requests with ones produced with the changes here, and it works properly. Notably, Llama3.1-8b-Instruct and Llama3.1-72b-Instruct do not get immediately aborted on imagegame anymore, since there is no extraneous || at the end of their cleaned responses anymore.

Testing result files will be removed before PR is put into review once this has been double-checked.

[Gnurro]

Testing data removed, ready for merge.