Add .col pronoun to all condformat rules (closes #19)

[zeehio]

Summary

Closes #19. Every CSS-based rule (rule_fill_discrete, rule_fill_gradient, rule_fill_gradient2, rule_fill_bar, rule_text_bold, rule_text_color, rule_css) evaluated its expression once and broadcast the result identically to every selected column, so a multi-column rule call couldn't express "each column's own condition" — only a shared one, or (with no expression) a warning-and-first-column-only fallback.

• Added a shared eval_expression_per_column() helper (rule_helper.R) that evaluates the expression once per selected column, with a .col pronoun bound in the data mask to that column's own values. Wired it into all seven rules, replacing each rule's "evaluate once, mat[, columns] <- value" pattern with a per-column loop that also recomputes any auto-detected limits/midpoint/palette from that column alone — matching the semantics of chaining the rule once per column:

  # These two are now equivalent:
  condformat(iris) %>% rule_fill_discrete(c(Sepal.Length, Sepal.Width), .col > 3)
  
  condformat(iris) %>%
    rule_fill_discrete(Sepal.Length, Sepal.Length > 3) %>%
    rule_fill_discrete(Sepal.Width, Sepal.Width > 3)

• expression now also defaults to .col when omitted, replacing the previous "applied to multiple columns, using column X ... please use an explicit expression" warning — each column now correctly uses its own values by default instead of silently only the first one.

• While rewriting rule_text_bold's per-column assignment, replaced its row-selector matrix indexing (bold_or_not_mat[bold_or_not, columns] <- "bold") with an ifelse()-based approach that can't hit R's "NAs are not allowed in subscripted assignments" error if the expression evaluates to NA for some rows; also dropped an unused, vestigial intermediate matrix in the same function.

A related, separate finding (not fixed here)

While touching rule_css.R, I noticed its na.value constructor argument is accepted, documented, and stored on the rule object, but rule_to_cf_field.rule_css never actually reads rule[["na.value"]] — so it's a dead parameter (unlike every other rule, where na.value is applied). Left as-is since it's an unrelated, pre-existing gap and a behavior change of its own; happy to fix in a follow-up if wanted.

Test plan

• Added a regression test per rule confirming the .col-based single-rule call produces output identical to chaining the rule once per column
• Updated the two existing tests that asserted the now-removed multi-column warning
• Updated roxygen docs, man pages (hand-synced since devtools::document() isn't available in this environment — worth a sanity check with devtools::document() before merge), and the introduction vignette
• CI runs green

---

Generated by Claude Code