fix(tests): resolve the emulator path before killing it

_emulator_process() returned the output of `ps -o comm=`, which on Linux is
the bare command name (`kkemu`, truncated to 15 chars), not a path -- only
macOS returns an absolute one. _power_cycle() then killed that pid and called
Popen([exe], cwd=cwd), and Popen resolves a bare name against PATH, never
against cwd. The emulator build directory is not on PATH, so on Linux the
emulator was killed and never restarted, leaving every later test in the run
talking to a dead port.

Resolve a runnable path BEFORE returning -- /proc/<pid>/exe first, then an
absolute comm, then cwd-relative, then PATH -- and report "not found" when
none works, so _power_cycle() takes its documented skip instead of killing an
emulator it cannot bring back.

Author: pastaghost

tests/test_msg_session_trust_lifetime.py

@@ -40,6 +40,7 @@
 from __future__ import print_function
 
 import os
+import shutil
 import subprocess
 import time
 import unittest
@@ -66,7 +67,9 @@
 TEST_KEY_ID = 3
 CI_SIGNER_ALIAS = 'CI Test'
 
-AAVE_V3_POOL = bytes.fromhex('7d2768de32b0b80b7a3454c06bdac94a69ddc7a9')
+# Aave V3 Pool proxy, matching AAVE_SUPPLY_SELECTOR below. Was the V2
+# LendingPool address, which does not expose supply().
+AAVE_V3_POOL = bytes.fromhex('87870Bca3F3fD6335C3F4ce8392D69350B4fA4E2')
 AAVE_SUPPLY_SELECTOR = bytes.fromhex('617ba037')
 PROBE_ARGS = [
     {'name': 'protocol', 'format': ARG_FORMAT_STRING, 'value': b'Aave V3'},
@@ -151,7 +154,45 @@ def _emulator_process(port):
             for cwd_line in cwd_out.splitlines():
                 if cwd_line.startswith('n'):
                     cwd = cwd_line[1:]
-            return pid, exe, cwd
+
+            # Resolve a RUNNABLE path before returning, because the caller
+            # kills this pid and then re-execs what we hand back.
+            #
+            # `ps -o comm=` gives the bare command name on Linux (`kkemu`,
+            # truncated to 15 chars), not a path -- only macOS returns an
+            # absolute one. Popen([name]) searches PATH, never cwd, and the
+            # emulator build directory is not on PATH. So on Linux the old
+            # code killed the emulator and then failed to restart it, leaving
+            # every later test in the run talking to a dead port.
+            #
+            # If no runnable path can be found, report "not found" so
+            # _power_cycle() takes its documented skip instead of killing an
+            # emulator it cannot bring back.
+            exe_path = _resolve_executable(pid, exe, cwd)
+            if exe_path is None:
+                continue
+            return pid, exe_path, cwd
+    return None
+
+
+def _resolve_executable(pid, comm, cwd):
+    """An absolute, runnable path for `comm`, or None."""
+    # Linux: the kernel knows exactly what is running.
+    try:
+        link = os.readlink('/proc/%d/exe' % pid)
+        if os.path.isfile(link) and os.access(link, os.X_OK):
+            return link
+    except (OSError, AttributeError):
+        pass
+    if os.path.isabs(comm) and os.access(comm, os.X_OK):
+        return comm
+    if cwd:
+        candidate = os.path.join(cwd, comm)
+        if os.path.isfile(candidate) and os.access(candidate, os.X_OK):
+            return candidate
+    found = shutil.which(comm)
+    if found:
+        return found
     return None