Implement maximum step limits and acceptance handling for slice sampling

[williamjameshandley]

This PR implements maximum iteration limits and acceptance handling for slice sampling to address infinite loop issues and improve robustness.

Changes

Core Implementation

• Add maximum step limits: Implement max_steps (stepping-out) and max_shrinkage (shrinking) parameters following Radford Neal's slice sampling algorithm
• Add acceptance handling: Include is_accepted boolean in SliceInfo and proper acceptance logic using static_binomial_sampling
• Prevent infinite loops: Bound both stepping-out and shrinking phases to avoid hanging

Parameter Improvements

• Better naming: Use max_steps and max_shrinkage for clarity
• Consistent defaults: max_steps=10, max_shrinkage=100 across both slice sampling and nested slice sampling
• Function organization: Rename internal functions to step_body_fun/step_cond_fun and shrink_body_fun/shrink_cond_fun

Nested Sampling Integration

• Full parameter support: Both NSS functions (build_kernel and as_top_level_api) now accept step limit parameters
• Consistent interface: Same parameter names and defaults across slice sampling and nested slice sampling modules

Issues Addressed

Closes #20 - Slice sampler infinite loop in horizontal_slice function

• Implements bounded stepping-out and shrinking phases
• Prevents indefinite loops that cause sampling sessions to hang

Closes #23 - Add max iterations limit to slice sampler while loops

• Adds configurable max_steps and max_shrinkage parameters
• Follows jaxopt pattern for bounded iterations with graceful failure handling

Closes #32 - Add acceptance boolean to slice sampling for handling failure modes

• Implements is_accepted boolean in SliceInfo with proper acceptance/rejection logic
• Uses static_binomial_sampling following BlackJAX patterns from other MCMC methods
• Provides graceful degradation when step limits are reached
• Implements maximum iteration limits for both stepping-out and shrinking phases

Implementation Details

Maximum Step Logic

• Stepping-out: Uses Neal's bounded approach where j and k counters limit expansion in each direction
• Shrinking: Limits iterations with s_steps < max_shrinkage + 1 condition to allow exactly max_shrinkage attempts
• Acceptance: Accepts proposals that find valid points within step limits, rejects otherwise

Backward Compatibility

• Default parameters maintain existing behavior for users not specifying limits
• All existing APIs preserved with new optional parameters
• Internal m parameter kept in horizontal_slice to maintain Neal's notation

Testing

This implementation has been tested to ensure:

• max_shrinkage=1 works correctly (previously failed)
• Step counting semantics are proper (iterations vs attempts)
• Acceptance logic properly handles boundary cases

🤖 Generated with Claude Code

Co-Authored-By: Claude [email protected]

[yallup]

This also in practice fixes #30 for the default delete function. Outer loop boilerplate needs adjusting to respect the return of a duplicated sample (failed constrained sample) as a signal for early termination