common: use enums for sampler types

[z80maniac]

This PR does some refactoring of the code related to working with sampler names. Its main purpose is to prepare the code for adding sampling-seq to the server.

1. Instead of raw characters representing a sampler, now enum is used. I think this makes the code more readable and type-safe.
2. Sampler string parsing is now split into two functions: string_split and sampler_types_from_names. It was done because the latter function will also make parsing the sampler array for the server much easier.
3. Moved all bindings between sampler type and its names in one place, e.g. from llama_sampling_order_print to sampler_type_to_name_string. It makes it less likely to miss something and also sampler_type_to_name_string may be used in the get_formated_generation for the server.
4. The help info for --samplers now shows the actual default value, not the hard-coded example.