Feat/optimizations stability

[Aatricks]

This pull request introduces several documentation and code improvements aimed at clarifying LightDiffusion-Next's optimization stack, improving extensibility for model function wrappers, and refining the behavior and documentation of various caching and batching optimizations. The most significant changes include a new implementation report in the documentation, enhanced explanations of caching strategies, improved handling of model function wrappers for optimization stacking, and refinements to conditional batching logic.

Documentation improvements:

• Added links and sections for an "Implemented Optimizations Report" in the documentation, providing a source-based breakdown of current optimizations and their implementation status. [1] [2] [3]
• Clarified and updated the documentation for WaveSpeed/DeepCache and First Block Cache (FBCache), emphasizing which optimizations are user-facing and which are groundwork for future or specialized use. Updated configuration and usage instructions accordingly. [1] [2] [3] [4] [5] [6] [7] [8]
• Updated prompt caching documentation to reflect current cache structure, behavior, and best practices. [1] [2] [3]
• Clarified support status for SageAttention and SpargeAttn on RTX 50 series GPUs in both the README and SageAttention documentation. [1] [2]

Codebase enhancements:

• Introduced ModelFunctionWrapperChain in ModelPatcher.py to allow multiple model function wrappers to be composed and applied in order, preventing silent disabling of earlier optimizations when stacking wrappers. Updated set_model_unet_function_wrapper to use this chain. [1] [2]
• Improved conditional batching logic in calc_cond_batch to make batched_cfg behavior more explicit and robust, with clearer separation of conditional and unconditional branches when batching is disabled. [1] [2]
• Updated and expanded unit tests to support new batching logic and to record batch sizes for verification.

Other changes:

• Minor clarification and rewording in advanced CFG optimizations documentation to better explain the effects and limitations of joint conditional/unconditional batching. [1] [2] [3]
• Small code cleanup: removed device capability check in sageattention_enabled() to reflect updated support for newer GPUs.