A workaround for b/74837748 (and b/143186860)

Background: once kokoro job finishes, the content of the entire workspace (/tmpfs/src) is rsynced back to the kokoro agent (from where the artifacts matching the pattern are uploaded). The issue is that after finishing the build, the grpc workspace can be very large (~tens of gigabytes: e.g. 1. it has all the intermediate files produced during the build 2. it has all the cloned submodules 3. many jobs perform multiple builds, each in a separate copy of the grpc repository, so 1 and 2 are included multiple times). In addition to that, the rsync from windows and macos workers can be quite slow.
As a result, a lot of time can be spent at the end of the job, simply rsyncing the entire workspace (out of which only a few files matching the pattern will be used as artifacts and 99.9% of data is copied for no reason).
In the past, we tried to work around this problem by invoking delete_nonartifacts.sh at the end of selected jobs, which helps a little bit, but has number of other problems:

• even running delete_nonartifact.sh can be quite slow (e.g. we recently discovered that for windows grpc_build_artifacts it takes tens of minutes :-( )
• if the jobs times out and gets cancelled by kokoro, the delete_nonartifacts.sh may clash with the rsync started by kokoro, which results in thousands of "file has vanished" warnings flooding the build log (and making the log very hard to read) - see b/143186860
• it is easy to forget invoking delete_nonartifacts.sh in the CI jobs, so this approach is hard to maintain. Also, when the delete_nonartifacts.sh gets added to the end of the CI script, one must take care to correctly propagate the exit code of the actual test that ran previously (so this approach is error prone).

So this PR is taking a different approach.

• at the very beginning of the CI script run, simply move the entire workspace (/tmpfs/src is used by kokoro) to a different location (I chose /tmpfs/altsrc) and restart the CI script there. Only /tmpfs/src is rsynced by kokoro, so any intermediate files produced under /tmpfs/altsrc won't cause any finalization overhead
• care must be taken to make sure that all the files we actually want to keep (the artifacts produced by the build and the tests reports that will be later displayed in the UI) will actually be stored under the original location (/tmpsfs/src), otherwise they won't be stored by kokoro.

Most of the logic of this script is:

• a move_src_and_respawn script (for unix and windows) that can be easily started at the beginning of any CI script and it takes care of the dirty job of switching the current workspace.
• make sure the test reports and artifacts end up in the right spot (note that the approach I chosed make sure that the reports end up in the right spot even if the test scripts crashes abruptly and there's no "copy afterward" steps for test reports).

This PR switches all the macos and windows CI jobs to use the "altsrc", since kokoro's rsync is much slower on macos and windows.
On linux I'm only switching some selected jobs to "altsrc", since 1. rsync quite fast on linux 2. most linux builds run under a docker container, which doesn't pollute the workspace with intermediate files (since the intermediate files are inside a container that is thrown away once the build finishes), so the workspace size on linux is generally smaller than on macos and windows.