add memory importance scoring via prompt logprobs

score_memories() drops each memory from the context one at a time,
runs prompt_logprobs against the full conversation, and builds a
divergence matrix: memories × responses.

Row sums = memory importance (for graph weight updates)
Column sums = response memory-dependence (training candidates)

Uses vLLM's prompt_logprobs to check "would the model have said
this without this memory?" — one forward pass per memory, all
responses scored at once. ~3s per memory on B200.

Co-Authored-By: Proof of Concept <poc@bcachefs.org>

src/thought/mod.rs

@@ -15,6 +15,7 @@ pub mod glob_tool;
 pub mod grep;
 pub mod memory;
 pub mod read;
+pub mod training;
 pub mod write;
 
 pub use bash::ProcessTracker;