Unfortunately, answers to these questions are not yet available. As Morgan et al (2016) put it:
... whether and to what extent science achievement gaps display
stable, cumulative, or compensatory growth trajectories are unknown.
This has been due to the lack of large-scale, multivariate, and
longitudinal studies, particularly those following diverse samples
across the elementary and middle school grades.
Available data (eg. the ECLS) typically follows children only as far as middle school, so effects on post-secondary level students are hard to come by.
That said, as the Elisabeth McClure reports listed in the question assert, the available data suggests that early STEM education has significant impact on academic achievement and cognitive skills later in life. The earlier the intervention, the greater the impact down the road, particularly for pre-kindergarten (eg, Duncan et al, 2007; Sackes et al, 2010; Tippett & Milford, 2017), and early kindergarten (eg, Sarama et al, 2012) exposure. That is, even where achievement gaps may be compensated for with later remedial programs, the sooner children are exposed to STEM education, the more favourable the future effects.
Correspondingly, interest in STEM education and careers is already set well before high school - from Sadler et al (2012):
The key factor predicting STEM career interest at the end of high
school was interest at the start of high school. ... During the high
school years, the percentage of males interested in a STEM career
remained stable (from 39.5 to 39.7), whereas for females it declined
from 15.7 to 12.7.