Fix cross-language inconsistencies; add print/fast parity

[Yicheng-Zhao1999]

Cross-language consistency fixes + feature parity (R)

Aligns R/gmd.R with the Python/Stata packages and adds two options those packages
already have. Verified against live version 2026_03.

Coverage (this package = R)

#	Finding	Needed in R?	Changed in this PR?
1	All-missing rows not dropped as documented	No (R already drops all-missing rows)	No
2	Variable names case-sensitive	Yes	Yes — case-insensitive + canonical casing
3	list/current/load keywords case-sensitive	No (R already case-insensitive)	No
4	Whitespace not trimmed	Yes (errored)	Yes — trim; empty/blank -> no filter
5	raw with a non-boolean value	No (errors on "yes"/"0"; "false" -> off)	No
6	sources + a variable that source lacks	No (R already errors)	No
7	Error precedence (version vs cite)	Yes (did not validate version before cite)	Yes — validates version first
—	print	Yes (missing)	Yes — added print_option
—	network	Missing, but intentionally not added	No (see below)
—	fast	Yes (missing)	Yes — added fast

What changed

• Update and fix the raw option bug #2 variables = "rgdp" matches and is normalized to canonical casing (rGDP).
• Update var list #4 country = " USA ", version = " 2026_03 ", whitespace-only / empty vectors no
  longer error; blank tokens are dropped and an empty value becomes NULL ("no filter").
• Robustness fixes for the R package (#318): version/input validation, case-insensitive variables, year filtering, httr2 #7 gmd(cite = "GMD", version = "2099_99") now errors on the invalid version
  instead of silently returning the citation.
• print print_option = "GMD"/"Stata" (case-insensitive) prints the APA citation
  and returns invisibly; text is byte-identical to the Python output.
• fast fast = TRUE caches the downloaded dataset under
  tools::R_user_dir("globalmacrodata", "cache"); the original .dta bytes are stored
  verbatim so a cached read is byte-identical to a fresh download.

Intentionally not added: network

Python/Stata's network overrides a connectivity heuristic: when their version fetch
fails they assume "offline", load a local dataset, and refuse sources/raw/cite unless
network="yes" forces them. R has no such global verdict -- it attempts each real
request independently and falls back / errors per resource. There is nothing for
network to override, so it would be a no-op in R.

Pending decision

• Return form of metadata listing. gmd(vars=TRUE) / gmd(iso=TRUE) return a
  data.frame in R, while Python/Stata print and return None (and Python exposes
  list_variables() / list_countries() for the data). So x <- gmd(vars=TRUE) yields
  data in R but None in Python. Left as-is pending a decision on whether to align.