Here are my findings.
Quite logically you need a non-buggy mit-scheme to build mit-scheme 10 from the git repo. The latest binary release (10.1.10) downloadable from the main site https://www.gnu.org/software/mit-scheme/ is affected by the bug. 10.1.9 isn't. It's available here: https://ftp.gnu.org/gnu/mit-scheme/stable.pkg/10.1.9/mit-scheme-10.1.9-x86-64.tar.gz
Once you have a clean binary, the branch `release-10' does build and is free from the vector bug.
git clone git://git.savannah.gnu.org/mit-scheme.git
cd mit-scheme
git checkout release-10
cd mit-scheme/src
./Setup.sh && ./configure && make
make install
The master branch doesn't build, but it's being actively worked on (the last commit was 43 hours ago), so nothing unusual here.
gcc -DHAVE_CONFIG_H -DMIT_SCHEME -DDEFAULT_LIBRARY_PATH=\"/usr/local/lib/mit-scheme-x86-64-10.90\" -I. -I. -O3 -frounding-math -fno-builtin-floor -Wall -Wclobbered -Wempty-body -Wignored-qualifiers -Wimplicit-fallthrough -Wmissing-field-initializers -Wmissing-parameter-type -Wnested-externs -Wold-style-declaration -Woverride-init -Wpointer-arith -Wredundant-decls -Wshift-negative-value -Wtype-limits -Wundef -Wuninitialized -Wwrite-strings -Wno-error=stringop-truncation -Werror -o syntax.o -c syntax.c
syntax.c: In function ‘Prim_scan_sexps_forward’:
syntax.c:1074:15: error: ‘level_start.last’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
1074 | (((level -> last) == NULL)
| ^~
syntax.c:1036:33: error: ‘level_start.last’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
1036 | (level -> previous) = (level -> last);
| ~~~~~~~^~~~~~~~
cc1: all warnings being treated as errors
make[2]: *** [Makefile:182: syntax.o] Error 1
make[2]: Leaving directory '/home/anon/mit-scheme/src/microcode'
make[1]: *** [Makefile:796: microcode/scheme] Error 2
make[1]: Leaving directory '/home/anon/mit-scheme/src'
make: *** [Makefile:699: all] Error 2
The fact that the release available from the MIT/GNU Scheme comes with such a bug and has stayed like that for 10 months is telling something about the number of users (plus all the linux distros that package this version) That's what should be really fixed. Maybe with a new binary build from the `release-10' branch?