- Added `logits_i` argument to `Context.ApplyPenalty`
 - Added a new exception type for `llama_decode` return code

- Added a `DecodeAsync` overload which runs the work in a task
 - Replaced some `NativeHandle` usage in `BatchedDecoding` with higher level equivalents.
 - Made the `LLamaBatch` grow when token capacity is exceeded, removing the need to manage token capacity externally.

- Made `NativeApi` into a `static class` (it's not intended to be instantiated)
 - Moved `LLamaTokenType` enum out into a separate file
 - Made `LLamaSeqId` and `LLamaPos` into `record struct`, convenient to have equality etc

- Added a `Sample` method to `LLamaContext` which uses a custom pipeline
 - Modified all executors to use the custom pipeline if it exists

- Cleaned up comments in implementations of `IInferenceParams`
 - Removed default values for all parameters in `LLamaContext.Sample` - they're never used and probably _shouldn't_ ever be used

had to update binary to `b1426`

- `AntipromptProcessor` accepts chunks of text and returns a value indicating if any antiprompt has been detected.
 - `StreamingTokenDecoder` decodes tokens into text, maintaining some internal state to handle single characters which are encoded as multiple tokens.

Added tests for these classes and updated StatelessExecutor to use them.

Removed most DeTokenize methods, marked the rest as obsolete (should always use a `StreamingTokenDecoder`).

- Built a new (hacky) `Detokenize` method which handles this

Major llama.cpp API Change

- params contains a list of loras, instead of just one

- Added a "spinner" to the `StatelessModeExecute` demo, which spins while waiting for the next token (demonstrating that it's not blocked).

better_instruct_antiprompt_checking

`GetState()` fix

- Just convert it to a `string`, nice and simple
 - Write the bytes to a `Span<byte>` no allocations
 - Write the chars to a `StringBuilder` potentially no allocations

- Fixed multiple enumeration
 - Fixed newline penalisation

ModelsParams record class

- Used those methods to add a `Clone` method to SafeLLamaContextHandle
 - Simplified `LLamaContext` by using the new methods
 - Sealed `LLamaContext` and `LLamaEmbedder`

- Passing encoding in the `IModelParams`, which reduces how often encoding needs to be passed around

- Removed some unnecessary `ToArray` calls
 - Initial pass on LLamaStatelessExecutor, the context overflow management is broken but I think I found where it's ported from

- Converted `SafeLLamaContextHandle` to take a `ReadOnlySpan` for Eval, narrower type better represents what's really needed

- Integrated grammar into sampling
 - Added a test for the grammar sampling

- sealed some classes not intended to be extended