fix: robust argument evaluation for NSE functions in loops

[laresbernardo]

Description

This PR fixes a critical bug introduced in version 5.2.8 that caused corr_var(), ci_var(), and balance_data() to crash when used in for loops, lapply(), or Shiny reactive environments where column references are parameterized (e.g., names(iris)[i], n_iris[i], etc.).

What was happening:

1. Data-Mask Collision: Previously, eval(substitute(var), df) was evaluated in the context of the data frame columns df. If the expression passed to var contained the dataset name (e.g., names(iris)[i] or iris[i]), R looked for a column named "iris" within iris first, resulting in evaluation failures.
2. Vector Evaluation Bug: If a character column vector itself (e.g., df[[5]]) was evaluated, is.character(var) would evaluate to TRUE. The function then mistakenly treated the entire character vector of length $N$ as the column name, causing downstream logic (like if (!var %in% colnames(rs$cor))) to crash with the condition has length > 1.
3. Broken ci_var() unquoting: In ci_var(), the unquoted string "Fare" inside summarise(mean(!!var, ...)) evaluated to the string literal "Fare" instead of the column symbol Fare, which returned NA values with warnings.

Proposed Changes:

1. Evaluate substitute(var) in parent.frame(): Evaluating in the parent environment prevents R from searching inside the data frame's columns for symbols like iris or df.
2. Strict Length & Type Checking: Added checks (is.character(var_eval) && length(var_eval) == 1 && !inherits(var_eval, "try-error")) to guarantee we only use the evaluated representation if it successfully resolves to a single string column name.
3. Correct Symbols in ci_var(): ci_var() now assigns sym(var_eval) (a symbol) or enquo(var) (a quosure), allowing it to properly unquote and evaluate inside tidyverse verbs.

---

Files Changed

• R/confidence.R: Robust parsing logic in ci_var(), returning a symbol/quosure for correct tidyverse unquoting.
• R/correlations.R: Robust parsing logic in corr_var().
• R/wrangling.R: Robust parsing logic in balance_data().

---

Verification Plan

Automated/Manual verification

Tested locally using devtools::load_all() with the following scenarios:

# 1. Loop test (AlexanderKlug's scenario)
n_iris <- names(iris)
for (i in 1:5) {
  print(corr_var(iris, var = n_iris[i], plot = FALSE))
}

# 2. ci_var bare name & string tests
ci_var(dft, Fare, Pclass)
ci_var(dft, "Fare")

# 3. balance_data bare name & string tests
balance_data(dft, Survived, rate = 1)
balance_data(dft, "Survived", rate = 0.5)

All of the above executed without errors and computed correct statistics/correlations.

Reported here: https://stackoverflow.com/questions/78437460/problem-with-assigning-variables-in-a-for-loop-with-lares-corr-var-function/78443283?noredirect=1#comment141107140_78443283