We should use humanize_number on the output size instead, so the JSON output can not humanize

Ahh, the trailing whitespace is coming from this.

You can just change this to:

xo_emit("{:description/%-8s/%s} ", e_desc.elm_desc_str, e_desc.elm_desc_str);

to skip the padding when outputing 'encoded' (json/xml etc)

We want the JSON output to have the size un-humanized, so you likely want this as something like:

xo_emit("{:size/%s/%zu}", size, mediasize);

This will use %s (size) when printing normally, and %zu (mediasize) when printing 'encoded' (JSON etc)

You'll need to change fetch_device_details to return the mediasize instead, then humanize the value into size for display.

In the "elements" array, we should add a 'type' field

Currently, you can't programmatically tell which things are disks and which are voltage/temperature sensors:

{
  "enc": "ses1",
  "name": "LSI SAS3x36 0601",
  "id": "500304801f14633f",
  "elements": [
    {
      "description": "Slot00  ",
      "device_names": "da24   ",
      "model": "HGST HUH721212AL4200    ",
      "serial": "8DGNJ3DH            ",
      "size": "11T"
    },
    {
      "description": "Slot01  ",
      "device_names": "da25   ",
      "model": "HGST HUH721212AL4200    ",
      "serial": "8DGNEJHH            ",
      "size": "11T"
    },
    {
      "description": "BackPlaneLM75",
      "temperature": 36
    },
    {
      "description": "ChipDie",
      "temperature": 71
    },
    {
      "description": "5V",
      "voltage": 5.00
    },
    {
      "description": "12V",
      "voltage": 11.90
    }
  ]
}

I've decided to left humanization to libxo(3) itself; this way we don't need to link with libutil. I've also changed it to match what's actually written on the disks.