Fix pytest result parsing counting more tests than were run

[KTanmay1]

Problem

RepoTransAgent/test_analyzer.py:126

test_pattern = r'(\S+)::\S+\s+(PASSED|FAILED|ERROR|SKIPPED)'
matches = re.findall(test_pattern, stdout)

\s matches newlines, so a test id ending one line pairs with a status token beginning the next. pytest's short summary emits exactly that shape for setup and teardown errors, whose lines carry no trailing reason:

ERROR tests/test_smaz.py::test_one
ERROR tests/test_smaz.py::test_two

Line 1 ends in ::test_one, line 2 begins with ERROR, and the pair matches. Chained across the block, N entries are counted as 2N - 1.

Observed on antirez_smaz (C → Python) against an empty tree: 18 tests run, 35 counted.

Reproduction

import re
pat = r'(\S+)::\S+\s+(PASSED|FAILED|ERROR|SKIPPED)'
out = """
tests/test_smaz.py::test_one ERROR                                       [ 33%]
tests/test_smaz.py::test_two ERROR                                       [ 66%]
tests/test_smaz.py::test_three ERROR                                     [100%]

=========================== short test summary info ============================
ERROR tests/test_smaz.py::test_one
ERROR tests/test_smaz.py::test_two
ERROR tests/test_smaz.py::test_three
============================== 3 errors in 0.02s ===============================
"""
print(len(re.findall(pat, out)))                 # 5  -- 3 tests ran
print(len(re.findall('^' + pat, out, re.M)))     # 3  -- correct

Scope

Only entries with no trailing reason chain:

• setup/teardown errors chain — ERROR tests/x.py::test_one is bare
• collection/import errors do not — they print ERROR tests/x.py, with no :: to match
• failures do not — FAILED tests/x.py::test_y - AssertionError: ... has text after the id

Because Ni inflates while Ti does not, APR and AMPR are depressed on affected tasks. A fully passing run is unaffected, so SR is unchanged.

Fix

Anchor to line start and restrict the separator to spaces and tabs.

Verification

Through PythonTestAnalyzer.analyze_output, not the regex alone:

input	before	after
3 setup errors	total=5	total=3, rate 0.0%
2 passed, 1 failed	total=3	total=3, rate 66.7%
2 passed	total=2	total=2, rate 100.0%

This changes reported APR/AMPR on tasks with setup or teardown errors. Flagging that explicitly since it affects published results — happy to hold if you would rather land it alongside a re-run.