Provide preset configs in library

[jerinphilip]

Currently, the Python calling code looks like:

config = Config()
model = Model(config, package)

This is not informative or self-contained enough of the intentions. This PR tries to change this to:

tiny: Config = slimt.preset.tiny()
model = Model(tiny, package)

base: Config = slimt.preset.base()
model = Model(base, package)

It's mostly cosmetic, with now the ability to indicate whether we're using a tiny configuration or a base configuration. This may also allow getting the borked deen base model to work. The data-members on Config are exposed for read-write via python.

Broken en-de-base

$ slimt translate -m en-de-base <<< "Hello world"
[warn] Failed to ingest expected load of Wemb_QuantMultA
[warn] Failed to ingest expected load of special:model.yml
[warn] Failed to complete expected load of decoder_l2_ffn_ffn_ln_bias
[warn] Failed to complete expected load of decoder_l2_rnn_ffn_ln_scale
[warn] Failed to complete expected load of decoder_l2_ffn_b2
[warn] Failed to complete expected load of decoder_l2_ffn_W2
[warn] Failed to complete expected load of decoder_l2_ffn_b1
[warn] Failed to complete expected load of decoder_l2_context_bo
[warn] Failed to complete expected load of decoder_l2_context_Wv_QuantMultA
[warn] Failed to complete expected load of decoder_l2_context_Wk_QuantMultA
[warn] Failed to complete expected load of decoder_l2_ffn_ffn_ln_scale
[warn] Failed to complete expected load of decoder_l2_context_Wk
[warn] Failed to complete expected load of decoder_l2_context_Wq_QuantMultA
[warn] Failed to complete expected load of decoder_l2_context_bq
[warn] Failed to complete expected load of decoder_l2_context_Wq
[warn] Failed to complete expected load of decoder_l2_context_Wo_QuantMultA
[warn] Failed to complete expected load of decoder_l2_context_Wo
[warn] Failed to complete expected load of decoder_l2_context_Wo_ln_scale
[warn] Failed to complete expected load of decoder_l2_ffn_W2_QuantMultA
[warn] Failed to complete expected load of decoder_l2_rnn_bf
[warn] Failed to complete expected load of decoder_l2_rnn_Wf
[warn] Failed to complete expected load of decoder_l2_context_bv
[warn] Failed to complete expected load of decoder_l2_rnn_W_QuantMultA
[warn] Failed to complete expected load of decoder_l2_rnn_ffn_ln_bias
[warn] Failed to complete expected load of decoder_l2_rnn_W
[warn] Failed to complete expected load of decoder_l2_ffn_W1_QuantMultA
[warn] Failed to complete expected load of decoder_l2_context_bk
[warn] Failed to complete expected load of decoder_l2_context_Wo_ln_bias
[warn] Failed to complete expected load of decoder_l2_context_Wv
[warn] Failed to complete expected load of decoder_l2_ffn_W1
[warn] Failed to complete expected load of decoder_l2_rnn_Wf_QuantMultA

Will probably need to figure out what is happening with the model, it does not work still - but at least the configurability is a net plus.