feat: Add from_borrowed() constructor

[zharinov]

Adds from_raw_parts() for constructing models from pre-parsed components, from_pretrained() now delegates to it.

Also fixes a bug where loading would fail if the tokenizer doesn't define an unk_token (not all tokenizers have one).

[Pringled]

Thanks for making this PR @zharinov! This is a nice functionality to have I think, and good catch about the unk_token. I have two small (but nice to have) improvements; if you could implement those this is good to go. Thanks for updating the tests as well 👍

[Pringled]

@zharinov one additional comment, could you also run clippy to fix the formatting issues?

[zharinov]

Hey,

I wanted to support zero-copy initialization with include_bytes!, yet missed the array clone in my own code 🤦‍♂️

The second attempt transforms StaticModel to use Cow, allowing for both owned and static scenarios. This adds some performance penalty, but hopefully it's minimal thanks to CPU branch prediction.

Also, I applied the suggestion for unk_token.

[zharinov]

UPD. Once I had benchmarks set up locally, I've done some additional research if you're interested. Here is the report I've got:

1. ndarray SIMD vectorization in pool_ids

// Before: manual loops
let mut sum = vec![0.0; dim];
for (i, &v) in row.iter().enumerate() {
    sum[i] += v * scale;
}

// After: ndarray vectorized ops
let mut sum = Array1::<f32>::zeros(dim);
sum.scaled_add(scale, &row);

2. .copied() pattern for hash lookups

// Before
*m.get(tok).unwrap_or(&tok)

// After
m.get(tok).copied().unwrap_or(tok)

3. AsRef<str> generics to avoid allocations

// Before
pub fn encode(&self, sentences: &[String]) -> Vec<Vec<f32>>

// After
pub fn encode<S: AsRef<str>>(&self, sentences: &[S]) -> Vec<Vec<f32>>

4. pool_ids takes slice instead of Vec

// Before
fn pool_ids(&self, ids: Vec<u32>) -> Vec<f32>

// After
fn pool_ids(&self, ids: &[u32], max_length: Option<usize>) -> Vec<f32>

Benchmark Results:

Benchmark	Improvement
encode_single/short	-8%
encode_single/medium	-8%
encode_single/long	-7%
encode_batch/100/short	-12%
encode_batch/100/medium	-16%
encode_batch/100/long	-16%

Branch: https://github.com/zharinov/model2vec-rs/tree/opt/all-optimizations