bpo-39151: Simplify dfs in the assembler

[pablogsal]

This simplification has the advantage that we don't stack overflow if the amount of jumps in the blocks is insane (maybe some autogenerated code).

The current code already relies on the topological order of the blocks, this PR just continues with that assumption to further simplify.

https://bugs.python.org/issue39151

[pablogsal]

As a bonus, this does hilariously jumpy code compilation faster:

master

./python.exe -m pyperf timeit 'code = ("\n".join("  "*i + "if x:\n" + "  "*i + "  ...\n" + "  "*i + "else:" for i in range(30)) + "\n" + "  "*30 + "...")' 'compile(code, "<dummy>", "exec")'
.....................
Mean +- std dev: 785 us +- 16 us

patched

./python.exe -m pyperf timeit 'code = ("\n".join("  "*i + "if x:\n" + "  "*i + "  ...\n" + "  "*i + "else:" for i in range(30)) + "\n" + "  "*30 + "...")' 'compile(code, "<dummy>", "exec")'
.....................
Mean +- std dev: 734 us +- 9 us

[csabella]

@pablogsal, should this be merged since it was approved?

[pablogsal]

Thanks, @csabella for the ping! :)