Remove dplyr and rlang dependencies (Phase 3 dependency reduction)

[chris-prener]

Summary

Remove dplyr and rlang by replacing all 37 call sites (17 dplyr + 20 rlang) with base R equivalents. This is the largest refactoring phase from the dependency reduction research (#36).

Scope

Not targeted for the next CRAN release. This is a longer-term improvement to pursue after initial CRAN acceptance.

Replacements

dplyr function	Base R equivalent	Call sites
mutate()	transform() / df$col <-	multiple
case_when()	ifelse() / vectorized logic	multiple
select() / rename()	bracket indexing + setNames()	multiple
left_join()	merge(x, y, all.x = TRUE)	multiple
`group_by()	> summarise()`	aggregate() / tapply()
all_of()	character vector indexing	multiple

rlang function	Base R equivalent
enquo() / sym() / quo_name()	match.arg() / string-based column selection

Design Considerations

• Removing rlang requires changing the programmatic column selection API pattern from tidy eval to string-based
• This may affect internal (non-exported) function signatures but must NOT change the public API
• tibble class output is lost unless another dependency provides it — functions will return data.frame instead
• Consider whether data.table is a lighter alternative (single dep replacing dplyr+rlang+tibble+tidyr)

Acceptance Criteria

• dplyr removed from DESCRIPTION Imports
• rlang removed from DESCRIPTION Imports
• All data manipulation rewritten in base R
• Tidy eval replaced with string-based column selection
• R CMD check passes with no new NOTEs
• All existing tests pass
• No public API behavior changes
• Performance benchmarked on representative datasets (no >2x regression)

Context

• Research: docs/requirements/dependency-reduction-recommendations.md
• Net result: 10 → 8 explicit Imports (after Phases 1 & 2)
• 7 files affected across R/

Epic

Part of #15 ([Epic A] CRAN Compliance & Release Readiness) — but deferred past initial CRAN submission

Epic

Part of #46 ([Epic E] Dependency Reduction)

[chris-prener]

Note: After removing dplyr, tibble will need to be re-added as an explicit Import to retain tibble-class output from public API functions. The net dependency change for Phase 3 is: remove dplyr + rlang, re-add tibble → 10 → 9 explicit Imports (not 8 as originally estimated).

[chris-prener]

API stability note: The public API should continue to accept bare (unquoted) column names — no breaking change to user-facing function signatures. Base R provides match.call() / substitute() / deparse() for non-standard evaluation (the same pattern library() uses to accept unquoted package names). These can replace rlang::enquo() / sym() / quo_name() while preserving the existing user-facing interface.