Log can succeed on distributions with probability mass less than 0

[Hazelfire]

Description:

log should fail on distributions with probability mass lower than 0, for instance log(normal(1, 1))

This actually causes issues further down the track, for instance, normalizing one of these invalid distributions ends up with an empty graph:

Steps to reproduce:

Using currently 0.2.2 on @quri/squiggle-components on a Squiggle notebook. Just pop the expressions above

Expected behavior:

I would prefer this to error when doing log(normal(1, 1)) rather than continue to that point and fail later.

What I got instead:

It succeeds / fails on normalizing the above expressions

[OAGr]

Good find. Makes sense.