Mitigate XS CI build flakes

[michaelfig]

I've found that occasionally the XS build crashes the GCC compiler, and returns an exit status 2. I can replicate it by running:

status=0; while test $status -eq 0; do
  rm -rf $MODDABLE/build/tmp/lin/release/eventual-send
  mcconfig -m -p x-cli-lin test-xs-manifest.json
  status=$?
done; echo "Final status=$status"

This loop runs normally for the most part, but eventually crashes the mcconfig compile.

expected identifier at end of input

``` /home/runner/moddable/build/tmp/lin/release/marshal/mc.xs.c:2748:12: error: expected identifier at end of input /* 1460 */ { (txSlot*)&gxHeap[1461], {.ID = XS_NO_ID, .flag = 0xf0, .kind = XS_INSTANCE_KIND}, .value = { .instance = { NULL, (txSlot*)&gxHeap[656] } } }, ^ [...] ##[error]Process completed with exit code 2. ```

excess elements in union initializer

``` /home/runner/moddable/build/tmp/lin/release/marshal/mc.xs.c:5059:0: warning: excess elements in union initializer /* 3771 */ { (txSlot*)&gxHeap[3772], {.ID = xsID_isNaN, .flag = 0x8e, .kind = XS_REFERENCE_KIND}, .value = { .reference = (txSlot*)&gxHeap[666] } }, [...] ##[error]Process completed with exit code 2. ```

error: ‘gxH’ undeclared here (not in a function)

``` /vagrant/moddable/build/tmp/lin/release/eventual-send/mc.xs.c:3711:121: error: ‘gxH’ undeclared here (not in a function) /* 2370 */ { NULL, {.ID = xsID_constructor, .flag = 0x8e, .kind = XS_REFERENCE_KIND}, .value = { .reference = (txSlot*)&gxHeap[2364] } }, ^~~ /vagrant/moddable/build/tmp/lin/release/eventual-send/mc.xs.c:3711:12: error: expected ‘}’ at end of input /* 2370 */ { NULL, {.ID = xsID_constructor, .flag = 0x8e, .kind = XS_REFERENCE_KIND}, .value = { .reference = (txSlot*)&gxHeap[2364] } }, ^ /vagrant/moddable/build/tmp/lin/release/eventual-send/makefile:227: recipe for target '/vagrant/moddable/build/tmp/lin/release/eventual-send/mc.xs.c.o' failed make: *** [/vagrant/moddable/build/tmp/lin/release/eventual-send/mc.xs.c.o] Error 1 make: *** Waiting for unfinished jobs.... Final status=2 ```

With some shell-scriptery, the CI build can be a little more tolerant of these errors, and rerun the mcconfig steps if they fail. That would help reduce the frequency of test flakes.

[dckc]

dup of #442 ?