Handle modern memory output from Arista EOS 4.27+

Arista EOS 4.27+ (and possibly earlier versions, untested on
4.24-4.26) changes the output from `show processes top once`
command, rendering values decimal numbers of MB instead of integer
numbers of KB, and also reorders the fields.

This commit extends the `get_environment` function with additional
logic to handle the output from both older and newer OS versions.

The memory expressed in MB is converted into KB for backward
compatibility, but because the switch has done some rounding and
displays only a limited number of decimal places, the conversion
into KB will not be exact. There is nothing that Napalm can do
about this.

Fixes #1671

napalm/eos/eos.py

@@ -846,16 +846,31 @@ def extract_temperature_data(data):
         # Matches either of
         # Mem:   3844356k total,  3763184k used,    81172k free,    16732k buffers ( 4.16 > )
         # KiB Mem:  32472080 total,  5697604 used, 26774476 free,   372052 buffers ( 4.16 < )
+        # MiB Mem :   3889.2 total,    150.3 free,   1104.5 used,   2634.4 buff/cache (4.27 < )
         mem_regex = (
-            r"[^\d]*(?P<total>\d+)[k\s]+total,"
-            r"\s+(?P<used>\d+)[k\s]+used,"
-            r"\s+(?P<free>\d+)[k\s]+free,.*"
+            r"^(?:(?P<unit>\S+)\s+)?Mem\s*:"
+            r"\s+(?P<total>[0-9.]+)[k\s]+total,"
+            r"(?:\s+(?P<used1>[0-9.]+)[k\s]+used,)?"
+            r"\s+(?P<free>[0-9.]+)[k\s]+free,"
+            r"(?:\s+(?P<used2>[0-9.]+)[k\s]+used,)?"
+            r".*"
         )
         m = re.match(mem_regex, cpu_lines[3])
-        environment_counters["memory"] = {
-            "available_ram": int(m.group("total")),
-            "used_ram": int(m.group("used")),
-        }
+
+        def _parse_memory(unit, total, used):
+            if unit == "MiB":
+                return {
+                    "available_ram": float(total) * 1024,
+                    "used_ram": float(used) * 1024,
+                }
+            return {
+                "available_ram": int(total),
+                "used_ram": int(used),
+            }
+
+        environment_counters["memory"] = _parse_memory(
+            m.group("unit"), m.group("total"), m.group("used1") or m.group("used2"))
+
         return environment_counters
 
     def _transform_lldp_capab(self, capabilities):