[ggdesplot] fix spurious colour legend, named-palette fallback, and multi-facet

[SchmidtPaul]

These are three independent fixes in ggdesplot(). The lattice desplot() is not affected by any of them - in each case desplot() already produces the correct result, and ggdesplot() now matches it.

All reprexes use agridat::besag.met / synthetic data only.

---

1. Spurious no_color legend

When text or num is supplied without col, an internal dummy factor (no_color) leaks in as a visible colour legend.

library(desplot)
data(besag.met, package = "agridat")
ggdesplot(besag.met, yield ~ col*row, text = gen)
#> before: a legend titled "no_color" with a single entry "1" appears
#>         above the yield colorbar
#> after:  only the yield colorbar is shown (matches desplot())

Cause: without col, the code sets col.string = "no_color"; data[[col.string]] <- factor(1) and then unconditionally adds scale_color_manual(). The colour key is now suppressed (guide = "none") unless col was actually supplied.

---

2. Named col.regions / col.text fallback leaves cells uncolored

With a partial named palette, the warning promises "falling back to positional matching", but unmatched levels are drawn grey.

library(desplot)
d <- expand.grid(col = 1:4, row = 1:4)
d$rep <- factor(paste0("R", rep(1:4, length.out = 16)))   # R1..R4
ggdesplot(d, rep ~ col*row, col.regions = c(R1 = "red", R2 = "blue"))
#> before: R3 and R4 cells are grey50 (NA) despite the "positional matching" warning
#> after:  all four levels are colored positionally (red, blue, red, blue),
#>         matching desplot()

Cause: rep(col.regions, length = fill.n) kept the vector names, so scale_fill_manual() still matched by name and dropped unmatched levels. Names are now stripped (as.vector()) so the documented positional fallback actually applies. Same fix for the col.text fallback.

---

3. Only the first conditioning variable was used

A formula with two conditioning variables silently facets on the first only; cells from the dropped factor overplot each other.

library(desplot)
d <- expand.grid(x = 1:3, row = 1:3, site = c("S1","S2"), rep = c("R1","R2"))
d$site <- factor(d$site); d$rep <- factor(d$rep)
d$y <- as.numeric(interaction(d$site, d$rep))            # constant within a panel

ggdesplot(d, y ~ x*row | site + rep)
#> before: 2 panels (site only), 18 tiles per panel -> rep overplotted
#> after:  4 panels (site x rep), 9 tiles per panel, matching desplot()

Cause: facet_wrap() used ff$cond[1] only. All conditioning variables are now combined into a single panel factor, so every conditioning variable is faceted. Single-variable formulas keep their original panel labels.

---

Tests for all three are added in tests/testthat/test_ggdesplot_fixes.R. R CMD check is clean (0 errors, 0 warnings).

Note: these were found during an LLM-assisted bug hunt of desplot.