The origin of the correlations between mass, morphology, quenched fraction, and formation history in galaxies is difficult to define, primarily due to the uncertainties in galaxy star-formation histories. Star-formation histories are better constrained for higher redshift galaxies, observed closer to their formation and quenching epochs. Here we use "non-parametric" star-formation histories and a nested sampling method to derive constraints on the formation and quenching timescales of quiescent galaxies at $0.7<z<2.5$. We model deep HST grism spectroscopy and photometry from the CLEAR (CANDELS Lyman$-\alpha$ Emission at Reionization) survey. The galaxy formation redshifts, $z_{50}$ (defined as the point where they had formed 50\% of their stellar mass) range from $z_{50}\sim 2$ (shortly prior to the observed epoch) up to $z_{50} \simeq 5-8$. \editone{We find that early formation redshifts are correlated with high stellar-mass surface densities, $\log \Sigma_1 / (M_\odot\ \mathrm{kpc}^{-2}) >$10.25, where $\Sigma_1$ is the stellar mass within 1~pkpc (proper kpc). Quiescent galaxies with the highest stellar-mass surface density, $\log\Sigma_1 / (M_\odot\ \mathrm{kpc}^{-2}) > 10.25$, } show a \textit{minimum} formation redshift: all such objects in our sample have $z_{50} > 2.9$. Quiescent galaxies with lower surface density, $\log \Sigma_1 / (M_\odot\ \mathrm{kpc}^{-2}) = 9.5 - 10.25$, show a range of formation epochs ($z_{50} \simeq 1.5 - 8$), implying these galaxies experienced a range of formation and assembly histories. We argue that the surface density threshold $\log\Sigma_1/(M_\odot\ \mathrm{kpc}^{-2})>10.25$ uniquely identifies galaxies that formed in the first few Gyr after the Big Bang, and we discuss the implications this has for galaxy formation models.