The code looks excellent.  Thank you Ovid and it looks like Adrian as well.

Now I will tear it apart.

I have two internal issues with the subplan branch.

Having to call finalize() on the child... it seems like it should just do that
when it gets destroyed.  Fortunately this is shielded from the user by the
subtest() method.  This bit of the docs is the red flag...

  If the child falls out of scope before C<finalize> is called, a failure
  diagnostic will be issued and the child is considered to have failed.

  No attempt to call methods on a child after C<finalize> is called is
  guaranteed to succeed.

[Special nit, I hate the throwback &foo syntax, modern Perl programmers
probably won't even know what it means, and much prefer foo() to indicate a
function call in the docs]

Which leads into issue #2: The child functionality feels like it should go
into a subclass (or role or whatever).  Though the subplan code is decidedly
non-invasive so its quite open to argument.

Externally, I have only one nit.  suite_passed() is a good idea, except...

$ perl -Ilib -wle 'use Test::More;  pass();  print "Yes" if
Test::More->builder->suite_passed'
ok 1
Yes
# Tests were run but no plan was declared and done_testing() was not seen.

For anything but a finalized child object it is a lie.  The fact that the
method name is past tense is the problem.  You can only know if the suite has
passed after all the testing is done.  Before that it can not be known and the
output is gibberish.  What you can know is if the suite is *currently
passing*.  That is, has anything happened up to this point to state that the
suite has failed?  This is useful information at any time in the life of a
builder object and its an oft requested feature.

The work done for suite_passed() can easily be subverted for is_passing()
which I will do now.