fix: prevent unbounded memory growth in parse_powermetrics

[zubvit]

Problem

powermetrics appends a null-byte-separated plist entry to its output file on every interval tick and never truncates it. parse_powermetrics called fp.read() on the entire file every loop iteration — after an hour at the default 1-second interval, it loads thousands of stale samples into RAM on every tick, with memory growing linearly with uptime.

Fix

Seek to 64 KB before EOF and read only the tail. A single plist sample is 4–16 KB, so data[-1] still gets the latest complete entry while memory usage becomes O(1) regardless of how long asitop runs.

-        with open(path+timecode, 'rb') as fp:
-            data = fp.read()
+        with open(path+timecode, 'rb') as fp:
+            fp.seek(0, 2)
+            size = fp.tell()
+            fp.seek(max(0, size - 65536))
+            data = fp.read()

Tested on macOS with Apple M1, asitop 0.0.24.