Jrmales/modal ps ds restart hardening

[jaredmales]

This work was performed by Codex (GPT-5) in response to the prompt: "after reviewing agent_context.md, please analyze the app modalPSDs (app/modalPSDs) looking for data races, restart bugs, memory management problems, etc. This app routinely crashes when the inputs resize or are recreated."

Summary

This PR adds a fixed-size SPSC specialization for mx::sigproc::circularBufferIndex to support low-latency, restart-safe producer/consumer handoff for modalPSDs.

The new specialization is intended for:

• fixed-capacity operation after maxEntries()
• one producer thread
• one consumer thread
• atomic publication of ring-buffer metadata
• snapshot/load-style access to coherent logical windows

Motivation

modalPSDs needed a way to keep its steady-state source-pointer buffering zero-copy and low-overhead while still avoiding the unsynchronized push_back()/growth behavior and metadata races present in the generic dynamic circular-buffer path.

This PR introduces a dedicated fixed-size SPSC mode rather than changing behavior for existing users of the legacy circular buffer.

Scope

The new specialization adds:

• fixed-size storage setup
• atomic latest / nextEntry / validEntries / mono publication state
• snapshot metadata describing the readable region
• validated sequence-load helpers for coherent consumer-side reads

Existing non-fixed-size users remain on the legacy implementation.

Validation

This change was validated through the downstream modalPSDs integration and test pass in MagAO-X.

Affected Files

• include/sigproc/circularBuffer.hpp

Follow-Up / Edge Cases

• The fixed-size specialization is intentionally SPSC-focused, not a general multi-producer/multi-consumer concurrent buffer.
• Additional direct mxlib unit coverage for the new specialization would still be worthwhile if this API is adopted more broadly.