Modified LLamaBatch to not share tokens with other sequences if logits is true. This ensures that the logit span at the end in used by exactly one sequence - therefore it's safe to mutate. This removes the need for copying _very_ large arrays (vocab size) and simplifies sampling pipelines.

* - Modified ISamplingPipeline to accept `ReadOnlySpan<float>` of logits directly. This moves responsibility to copy the logits into the pipeline.
 - Added a flag to `BaseSamplingPipeline` indicating if a logit copy is necessary. Skipping it in most cases.

* Fixed `RestoreProtectedTokens` not working if logit processing is skipped

* - Implemented a new greedy sampling pipeline (always sample most likely token)
 - Moved `Grammar` into `BaseSamplingPipeline`
 - Removed "protected tokens" concept from `BaseSamplingPipeline`. Was introducing a lot of incidental complexity.
 - Implemented newline logit save/restore in `DefaultSamplingPipeline` (only place protected tokens was used)

* Implemented pipelines for mirostat v1 and v2

Conversations can be "forked", to create a copy of a conversation at a given point. This allows e.g. prompting a conversation with a system prefix just once and then forking it again and again for each individual conversation. Conversations can also be "rewound" to an earlier state.

Added two new examples, demonstrating forking and rewinding.

- Added BaseSamplingPipeline which provides a base impl of `ISamplingPipeline`
 - Added `DefaultSamplingPipeline` which mimics normal llama.cpp sampling