Dsxir.Optimizer.SIMBA.Strategy.AppendDemo (dsxir v0.5.0)

SIMBA strategy that appends one augmented demo per predictor from the top trajectory in a bucket. No LM call.

Keeps the last demo per predictor name when a trace visits the same predictor more than once (DSPy parity: name2demo[name] = demo overwrites on repeat, then one demo per predictor is appended). Skips when the top trajectory score is at or below the mini-batch 10th percentile.

Input field values whose to_string length exceeds ctx.demo_input_field_maxlen are sliced and suffixed with "\n\t\t... <TRUNCATED FOR BREVITY>".