As Mark Twain said there are lies, damned lies, and statistics, and you are well within bounds to challenge exit polling data, especially when it's within the margin of error. I think it's safe to say Trump did do better with black people in 2020 than 2016, but I'm not sure how his performance compares with Bush 2004. The easiest way to answer this would be to look at the data in historically black areas and compare the results. Someone with a paid membership on Atlas could help here.
Yeah, the answer might be tough to say. This also wouldn't necessarily answer the question because African-Americans who vote Republican are disproportionately likely
not to live in overwhelmingly African-American areas, which have always been more Democratic than the black vote as a whole. We'd also need to consider measures of residential segregation, and just how representative strongly black areas are of the black population as a whole.
(Jews are the reverse; strongly Jewish communities are usually either ultra-Orthodox or have lots of recent Jewish immigrants, and are likely to vote comparatively Republican, but more assimilated Jews who don't live in strongly Jewish areas tend to be
strongly Democratic, such that the demographic as a whole is much more Democratic than you might initially guess. Native American reservations also vote much more Democratic than the Native American vote as a whole -- the population of people who want to live in strongly-
X communities, and people who don't, are going to differ in consistent ways for any given
X.)