Residual foreground contamination in cosmic microwave background (CMB) maps,
such as the residual contamination from thermal Sunyaev-Zeldovich (SZ) effect
in the direction of galaxy clusters, can bias the cross-correlation
measurements between CMB and large-scale structure optical surveys. It is thus
essential to quantify those residuals and, if possible, to null out SZ cluster
residuals in CMB maps. We quantify for the first time the amount of SZ cluster
contamination in the released Planck 2015 CMB maps through (i) the stacking of
CMB maps in the direction of the clusters, and (ii) the computation of
cross-correlation power spectra between CMB maps and the SDSS-IV large-scale
structure data. Our cross-power spectrum analysis yields a $30\sigma$ detection
at the cluster scale ($\ell=1500-2500$) and a $39\sigma$ detection on larger
scales ($\ell=500-1500$) due to clustering of SZ clusters, giving an overall
$54\sigma$ detection of SZ cluster residuals in the Planck CMB maps. The Planck
2015 NILC CMB map is shown to have $44\pm4\%$ of thermal SZ foreground emission
left in it. Using the 'Constrained ILC' component separation technique, we
construct an alternative Planck CMB map, the 2D-ILC map, which is shown to have
negligible SZ contamination, at the cost of being slightly more contaminated by
Galactic foregrounds and noise. We also discuss the impact of the SZ residuals
in CMB maps on the measurement of the ISW effect, which is shown to be
negligible based on our analysis.