fix(gatsby): merge data deps state instead of replaying actions

[vladar]

Description

This PR is a follow-up to #32305 After benchmarking on a heavy site we've learned that replaying actions for data dependencies is expensive. So basically this PR implements solution described in #32305 (comment):

If data dependencies will be too expensive to sync this way - we can do a hybrid approach - forward/replay some actions and merge expensive parts all at once in the end.

Now we merge the data dependency state at the end of the build. It is still expensive but far less expensive than replaying all data dependency actions: 21s vs 250s for this heavy test site:

success run queries in workers - 378.129s - 25789/25789 68.20/s
success Merge worker state - 21.678s

But query state files are huge, so maybe this is not a typical scenario (requires additional research):

-rw-r--r--  1 root root 320M Jul 20 15:38 redux.worker.slices_1_5d05d7d9a7164ada5d792186cc7691ab
-rw-r--r--  1 root root 306M Jul 20 15:38 redux.worker.slices_2_5d05d7d9a7164ada5d792186cc7691ab
-rw-r--r--  1 root root 317M Jul 20 15:38 redux.worker.slices_3_5d05d7d9a7164ada5d792186cc7691ab

[ch34739]

[pieh]

Looks good. My comments are not blocking and PR can be merged without addressing them (they could be addressed in follow up or just skipped completely) - up to you :)