Handle NaN likelihoods by treating as -inf while preserving NaN for user

[williamjameshandley]

Priority: Medium

Issue Description:
The slice sampler can get stuck when encountering NaN values in likelihood calculations. This contributes to the infinite loop problem reported in #20.

Proposed by: @williamjameshandley (Will Handley)
Context: Related to debugging stuck runs with @zwei-beiner (Namu Kroupa) and @yallup (David Yallup)

Proposed Solution:

• Treat NaN likelihood values as -inf internally for sampler logic
• Preserve the original NaN value for user purposes (logging, debugging, etc.)
• This prevents the sampler from hanging on NaN values while still allowing users to detect when their likelihood function produces NaN

Implementation Details:

• Add NaN detection in likelihood evaluation
• Convert NaN to -inf for internal sampling decisions
• Maintain separate record of original NaN values for user access
• Consider adding warning/logging when NaN values are encountered

Benefits:

1. Robustness: Prevents sampler hangs due to NaN values
2. Debugging: Users can still identify when their likelihood produces NaN
3. Continuity: Sampling can continue even with occasional NaN evaluations
4. Transparency: Clear handling of edge cases in likelihood evaluation

Related Issues:

• Slice sampler infinite loop in horizontal_slice function #20 Slice sampler infinite loop - directly addresses one suspected cause
• Add state saving/loading functionality for debugging stuck runs #21 State saving functionality - complements debugging capabilities
• Add max iterations limit to slice sampler while loops #23 Max iterations limit - complementary robustness improvements