is_pentagon() is broken for ISEA aperture-7 grids

[gcol33]

is_pentagon() is broken for ISEA aperture-7 grids — undercounts at resolution 1, hard errors at resolution >= 2

Defect

R/hex_topology.R:58-63:

pentagon_ids <- cpp_quad_ij_to_cell(
  quad = 0:11, i = rep(0, 12), j = rep(0, 12),
  resolution = res, aperture = ap
)
cell_id %in% pentagon_ids

Reproduction

Verified by running is_pentagon() across all cells from grid_global() at aperture 3/4/7, resolutions 1-3:

• Aperture 3 and 4: correctly report 12 pentagons at every resolution tested.
• Aperture 7, resolution 1: reports only 9 pentagons instead of 12. The candidate IDs from cpp_quad_ij_to_cell(0:11, 0, 0, res=1, ap=7) are 31 32 113 194 275 356 437 518 599 680 761 842, but the valid cell-ID range for that grid is only 30-789 (95 cells total) — several candidate IDs fall outside the real range, so %in% silently fails to match them.
• Aperture 7, resolution 2: throws Error: !anyNA(x) is not TRUE.
• Aperture 7, resolution 3: throws Error: cell_id must be a finite value in [1, 12252] for resolution 3, aperture 7.

For any aperture-7 ISEA grid at resolution >= 2, is_pentagon() is completely unusable (hard error); at resolution 1 it silently gives a wrong answer.

Impact

tests/testthat/test-topology.R only asserts the pentagon count for H3 grids (is_pentagon detects 12 pentagons for H3); there is no equivalent ISEA assertion, so this went uncaught.

Fix

The candidate-pentagon-ID computation needs to account for aperture-7's actual quad/i/j-to-cell numbering at each resolution rather than assuming (quad, 0, 0) maps directly for every aperture.

[gcol33]

Fixed in commit 17011c7. Root cause: candidate pentagon cell IDs were derived via a forward (quad,i,j)->cell_id computation that's unreliable for aperture 7 at low resolution (the true valid cell-ID range doesn't follow the same formula as apertures 3/4, so the computed candidates fell outside it or the encoding silently produced the wrong count). Rewrote is_pentagon() to decode each input cell ID's own (i,j) via the inverse direction and check for (0,0) directly, which sidesteps the unreliable forward path entirely. Verified exactly 12 pentagons found via full brute-force sweep of every valid cell ID for apertures 3/4/7 at resolutions 1-4 (was undercounting or throwing for ap7).