#276 diagnostics: STRATA_DUMP_FIRST_LOGITS writes the first generated token's logits

描述

## Summary

`STRATA_DUMP_FIRST_LOGITS=<file>` writes the logits of the token the prompt predicts (n_vocab float32, once per run).
Two variants of the engine run on the same prompt then give a first-token KL divergence: how far a change moves the
model's distribution, not only whether its top token changed. On a layer split the last stage's verifier writes it.

This is the measurement behind the precision numbers in the PRs that follow it (#280, #283,
#293, ...). These changes come from an NVFP4 fork of Strata (sergqwer/strata-nvfp4), ported onto current main one topic per PR.

## Correctness

Off unless the variable is set; nothing else changes.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

https://claude.ai/code/session_01VZy1yKaDDiA8a7svdwaHio