tokei seems to be dramatically mis-estimating Go code

[masklinn]

I was looking at https://go.dev/src/net/http/server.go in a local copy, using tokei 12.1.2 reports

===============================================================================
 Language            Files        Lines         Code     Comments       Blanks
===============================================================================
 Go                      1         3911         2906          764          241
===============================================================================
 Total                   1         3911         2906          764          241
===============================================================================

This seems credible, however while the extreme ratio of comments to code in the early part of the file does dry up some further down the file it never actually disappears. And indeed, using some basic rg filters for classification reveals a very different picture:

> rg '^$' src/net/http/server.go | wc -l # blank lines
     387
> rg '^\s*//' src/net/http/server.go | wc -l # comments
    1340
> rg -v '^(\s*//|$)' src/net/http/server.go | wc -l # code
    2184

I'm sure tokei has a smarter model than trivial regex, but the discrepancies seem really extreme, especially for a language which should be relatively easy to parse and classify. For what that's worth, loc returns the same things as the rg/wc values.