sf::st_transform() returns empty geometry

I have transformed the rnaturalearth countries dataset for orthographic plotting using the procedure outlined here.

However, transforming with st_transform() results in at least one of the geometries (mainland Russia) being empty.

This reads and attempts to transform the problematic polygon:

polygon <- st_as_sf(data.frame(st_as_sfc(readLines("https://pastebin.com/raw/APH15G6X"), crs = 4326)))
st_transform(polygon, "+proj=ortho +lat_0=20 +lon_0=0 +x_0=0 +y_0=0 +a=6371000 +b=6371000 +units=m +no_defs")

Simple feature collection with 1 feature and 0 fields (with 1 geometry empty)
Geometry type: MULTIPOLYGON
Dimension:     XY
Bounding box:  xmin: NA ymin: NA xmax: NA ymax: NA
CRS:           +proj=ortho +lat_0=20 +lon_0=0 +x_0=0 +y_0=0 +a=6371000 +b=6371000 +units=m +no_defs
            geometry
1 MULTIPOLYGON EMPTY

The geometry seems to be valid and is expected to be fully visible after orthographic projection, so I'm not sure what else could be going on.

The gist you linked is inspiring, and it has helped me a lot in the past. But it is rooted in an obsolete version of {sf} - and with release 1.0 (which introduces S2 backend for geographic CRS) things got somewhat easier.

I have updated the logic somewhat and published it a while back in this answer https://stackoverflow.com/a/70756593/7756889 - when I amend the code to your definition of Ortho projection it shows mainland Russia as expected (= the geometry is not empty, but cut somewhat to hide the invisible parts in Asia). In addition the answer uses the world dataset from Gisco instead of Natural Earth; it should have negligible impact.

library(sf)
library(giscoR) # for the countries dataset only
library(ggplot2)

# projection string used for the polygons & ocean background
crs_string <- "+proj=ortho +lat_0=20 +lon_0=0 +x_0=0 +y_0=0 +a=6371000 +b=6371000 +units=m +no_defs"

# background for the globe - center buffered by earth radius
ocean <- st_point(x = c(0,0)) %>%
  st_buffer(dist = 6371000) %>%
  st_sfc(crs = crs_string)

# country polygons, cut to size
world <- gisco_countries %>% 
  st_intersection(ocean %>% st_transform(4326)) %>% # select visible area only
  st_transform(crs = crs_string) # reproject to ortho

# now the action!
ggplot(data = world) +
  geom_sf(data = ocean, fill = "aliceblue", color = NA) + # background first
  geom_sf(fill = "lightyellow", lwd = .1) + # now land over the oceans
  theme_void()