Re. sampling:
You are absolutely correct to be conscious of the limitations of convenience samples. The CDC is also engaged in the more rigorous work of community level seroprevalence survey collection which uses systemic sampling, but those will take longer … they are not conveniently on hand.
That said in terms of antibody positive numbers/confirmed infection numbers my WAG is that any selection bias would actually go the other way in this case, for exactly the reason you stated, that these people may be more cautious than the general population overall because of underlying health conditions, so therefore less likely to be exposed, and also presumptively less likely to have remained asymptomatic or mildly symptomatic when infected. That’s the opposite than the convenience sampling of people out and about more on the street or at the grocery store.
We can also look to other countries. (Did here.) Spain had about 0.5% of the population with reported confirmed infections and 5% antibody positivity, so 10x. Sweden was running 59x more. The U.K. mid-April was up at 99x more (but had run over 30% positivity rate).
Re. Date Range:
A reasonable point which can be further supported. The key metric for apples to apples is the positivity rate. Positivity rates in that date range across the country were running 11 to 20%; now they are running 8 to 9%. The generally accepted principle is that higher positivity rates means more that were never diagnosed.
Then however there is the issue of who is getting infected now, the shift to a much younger population, who tend to have mild to asymptomatic infections much more commonly. It is reasonable to suspect that the tested group is barely scratching the surface of the numbers of these individuals who have no to minimal symptoms.
But then there is the bigger wrinkle: point prevalence of antibody positivity may dramatically underestimate past infections. We know that some number had antibodies which declined over time (which does not mean they are not, or are, immune), and some had infections with strong specific T-cell responses without ever having had an antibody response detected by the tools used. Small studies done so far (much harder to do than antibody ones) but so far they suggest that antibody positive results may miss at least half of those with even fairly recent infections.
So could the actual number be only 4 to 5x? Sure. Could it be 20 to 30x? Sure.