Fix two bugs in MCMC start_from feature

[Copilot]

The new start_from feature in the MCMC sampler contained two bugs that would cause runtime failures in specific scenarios.

Bugs Fixed

• collection.append() → collection._append(): SampleCollection exposes only a private _append() method. The public append() call would raise AttributeError in any MPI run where multiple chain files are present and need to be concatenated.

• max_tries.value accessed before NumberWithUnits conversion: _load_start_from() was called before the block that converts max_tries from its raw YAML value (e.g. "40d") into a NumberWithUnits object. Accessing .value on a bare string would raise AttributeError when sampled params are missing from the loaded chain. Fixed by:

  1. Moving _load_start_from() to after the NumberWithUnits conversion block
  2. Adding self.max_tries.set_scale(self.model.prior.d()) before using .value in the missing-params path — consistent with how normal initialization handles it

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.