Keeping model lists in sync, clarify `make` behavior (docs)

[mrienstra]

While working on #1806, I came across the following:

1. The list of models in ./models/README.md is currently a subset of the list of models in ./models/download-ggml-model.sh, which is itself a subset of what is available from https://huggingface.co/ggerganov/whisper.cpp/tree/main (bulk of options), https://ggml.ggerganov.com/ (duplicates of first only? Didn't check carefully), and https://huggingface.co/akashmjn/tinydiarize-whisper.cpp/tree/main (just one). Is there some logic to as to which models are included / not included? I was tempted to add more to both lists, but wanted to ask first about inclusion criteria.

2. In ./README.md#more-audio-samples, there's a list of make commands starting with make tiny.en, perhaps this used to be in a different section higher up? Are they still current / helpful? Oh, huh, I probably should've also changed this bit (in the "Quick start" section):

   Now build the main example and transcribe an audio file like this:

   # build the main example
   make
   
   # transcribe an audio file
   ./main -f samples/jfk.wav

   ... to clarify that make alone defaults to the base model, and also downloads the model if necessary (mentioned lower down: "The command downloads the base.en model converted to custom ggml format and runs the inference on all .wav samples in the folder samples."), meaning there's no need to run bash ./models/download-ggml-model.sh base.en if you just want to try the base model. Maybe I'll take another pass at improving the "Quick start" section after #1806 merges (or is closed). For someone who wants to just dive in, seems a little confusing to have basic useful info about make after a bunch of console output.

3. The list showing memory usage (./README.md#memory-usage) only shows 5 models, should there be an issue about bringing this up-to-date? Or at least adding -v1 to large, if that's the case?

4. The list of models in ./models/README.md gives SHA-1 file hashes, which is nice because they aren't long (keeps the table tidy), but also a smidge inconvenient, since Hugging Face provides SHA-256 files hashes (e.g. for large-v3). I suppose this could be considered a bonus, as one could -- after downloading a model -- compute locally & compare both values from both sources (./models/README.md & Hugging Face) for addition peace of mind / security.

5. ./Makefile doesn't seem to support all models, only those it is explicitly aware of, e.g. it doesn't know about large-v3-q5_0 or small.en-tdrz, so it fails with make: *** No rule to make target large-v3-q5_0'. Stop.So the list of models in./Makefileshould be keep somewhat in sync with e.g../models/download-ggml-model.sh` (see point 1)? Maybe a single source of truth would be nice?