saydo 0.1.0

• First release, built on the Neudata R package template.
• Renamed from llmCoder, which clashed with the unrelated llmcoder package on CRAN. The new name describes what the package measures: the gap between what people say and what they do. If you installed the GitHub version, reinstall with remotes::install_github("neu-data/saydo"), and note that the attributes stored on coded data are now saydo_codebook and saydo_id.
• llm_auto_code() applies a codebook to text with any ‘ellmer’ chat, or a custom coder function. Each text is coded in its own conversation, and the first request is sent alone so key, credit or model errors fail fast.
• code_agreement() compares LLM codes with human codes: percent agreement, Cohen’s kappa with a confidence interval, precision, recall and F1.
• cross_tab_themes() tests whether a numeric metric differs when each theme is present (Welch’s t-test, Benjamini-Hochberg adjustment), and can aggregate segment-level codes to participants before joining metrics.
• plot_theme_impact() plots the metric difference per theme with confidence intervals, in Neudata colours.
• mock_mixed_data() and mock_coder() let the whole workflow run offline.