cd_extract(): an area on sea cells only returns NaN without a warning #153

Problem

cd_extract() returns NaN with no warning for an area that falls entirely on ERA5-Land's sea cells. ERA5-Land has no values over the ocean (10,835 of the 31,581 cells in the live tmean_annual COG are NA), and cd_crop() (terra::crop(snap = "out", mask = TRUE), R/cd_crop.R:36) followed by global(mean, na.rm = TRUE) (R/cd_extract.R:77) gives NaN when every cell is NA.

Measured on the live catalog: a 0.02° polygon on Triangle Island (-129.08, 50.86) returns NaN for every year, silently. A 0.01° polygon inland near Squamish returns a value, because a polygon smaller than a cell picks up the cell it touches.

cd_extract_daily() already handles the point case: a point in a sea cell moves to the nearest land cell and is flagged cell_moved.

Proposed

Warn, naming the area, when an extract has no land cells; or follow cd_extract_daily() and take the nearest land cell, flagged. Found while reviewing the README tagline for #142.