feat(data): rebalance text conditioning toward structured Poucher profiles

feat(data): wire structured Poucher profiles into text conditioning pipeline

fix(training): update hf_train_job to run from cloned repo, add text conditioning, fix dataset viewer