fix(phylo): decode Gblocks stdout before regex match in run_gblocks

[wyim-pgl]

Problem

python -m jcvi.apps.phylo build crashes on Python 3 when Gblocks is enabled:

Traceback (most recent call last):
  File "jcvi/apps/phylo.py", line 921, in build
    gb_fasta = run_gblocks(mrtrans_fasta)
  File "jcvi/apps/phylo.py", line 216, in run_gblocks
    alignp = re.sub(
TypeError: cannot use a string pattern on a bytes-like object

Root cause

run_gblocks() applies a str regex to r, but r comes from AbstractCommandline.run() (in jcvi/compara/ks.py), which returns Popen(...).communicate(). Since jcvi.apps.base.Popen does not set text=True/universal_newlines, communicate() returns bytes on Python 3 — a leftover from the Python 2 port. This is independent of the Gblocks version.

Fix

Decode the captured stdout defensively before matching:

r = r.decode("utf-8", errors="replace") if isinstance(r, bytes) else r

Verification

• python -m py_compile src/jcvi/apps/phylo.py passes
• Simulated Gblocks 0.91b stdout (Gblocks alignment: 1234 positions (57 %)) as bytes now parses to alignp = 57; the str passthrough branch also works.

Note: the same bytes/str pattern exists in jcvi/compara/ks.py find_synonymous() (e.read().find(...) on a communicate() result), which will need a separate fix.