fix: report a missing git instead of an unhandled FileNotFoundError

[Eljees]

Closes #303.

Problem

With git not installed (or not on PATH), both entry points end in a traceback:

Traceback (most recent call last):
  ...
  File ".../diff_cover/command_runner.py", line 27, in execute
    with subprocess.Popen(command, stdout=stdout_pipe, stderr=stdout_pipe) as process:
FileNotFoundError: [Errno 2] No such file or directory: 'git'

Two things are going on:

1. command_runner.execute puts subprocess.Popen(...) outside the try. The
   try only covers process.communicate(), so the FileNotFoundError that Popen
   raises when the executable does not exist escapes untouched.

2. Both main() functions call GitPathTool.set_cwd(directory) before their own
   error handling starts. diff_quality_tool.main already has

   except OSError as exc:
       LOGGER.error("Failure: '%s'", str(exc))
       return 1

   and FileNotFoundError is an OSError — the handler for this case is already
   written, the failure just happens a few lines above the block that would catch it.
   diff_cover_tool.main has no handling at that point at all.

Precedent

This is the same class of defect as #378 ("Running diff-quality with a tool which is
not installed gives FileNotFoundError rather than helpful message"), fixed by #380.
That fix landed in run_command_for_code and covers "the linter is not installed";
the "git is not installed" path goes through execute and was left as it was.

Fix

• ExecutableNotFoundError(CommandError) — a subclass, so git_diff.py's
  except CommandError and every other existing handler keep working unchanged.
• execute() raises it for the FileNotFoundError from Popen only, with the message
  Improve error reporting when "git" is not found #303 asked for. The git-scm.com link is added only when the missing binary is git.
• Both main() functions wrap set_cwd and report through LOGGER.error, returning 1.

The narrow catch is deliberate

My first version caught CommandError wholesale in main(), and two of your existing
tests rejected it —
tests/test_integration.py::TestDiffCoverIntegration::test_git_diff_error and
::TestDiffQualityIntegration::test_git_diff_error_diff_quality. Both set up a git that
runs and exits 1, and both require CommandError to propagate out of main(). That is
the project saying "the command ran and failed" and "there is no command to run" are
different things, which is where the separate subclass came from. Both tests pass here.

Tests

tests/test_command_runner_missing_executable.py, six cases: execute(), a non-git
binary getting no git link, the subclass relationship, both main() functions, and

def test_a_command_that_runs_and_fails_is_left_alone(failing_git):
    """Control. Green before and after: only a *missing* binary is reworded."""

which is green on both sides — that is what shows the error semantics did not move.

Verification

	result
main (e34eec3, release 10.5.0), whole suite	382 passed, 0 failed
probe against the existing API before the fix	3 defects out of 3 — raw FileNotFoundError escaped execute() and both main()s
control before the fix (git present but failing)	handled as CommandError — silent, as it must be
this branch, whole suite	388 passed, 0 failed
new failures against main	none
reproducer from the issue, after the fix	one line of text, exit code 1, no traceback
blocking lint (black, isort, doc8)	clean

ruff and pylint are continue-on-error: true in verify.yaml, so I checked the
blocking three; diff-quality on this diff reports 100% for pyflakes.