Large refactor. Use rayon, 10X performance boost

Code changes:
Removed ignore & channel crates. Using a single reciever thread to build
a hashmap to prevend duplicate inodes being reported gave a severe
performance penalty

Using rayon crate with some hand crafted file traversal has improved
performance aprox 10X

Behaviour changes:

Removed parameter 'limit by filesystem' - don't think this is used, and
I only added it as it was easy to add with the ignore crate.

Sym links will now not appear in the output tree unless using '-s'
'apparent-size' flag

Change behaviour of multiple args so that it unifies them and
compares them under one tree instead of treating them
individually: https://github.com/bootandy/dust/issues/136

src/display_node.rs

@@ -0,0 +1,46 @@
+use std::cmp::Ordering;
+use std::path::PathBuf;
+
+#[derive(Debug, Eq, Clone)]
+pub struct DisplayNode {
+    pub name: PathBuf, //todo: consider moving to a string?
+    pub size: u64,
+    pub children: Vec<DisplayNode>,
+}
+
+impl Ord for DisplayNode {
+    fn cmp(&self, other: &Self) -> Ordering {
+        if self.size == other.size {
+            self.name.cmp(&other.name)
+        } else {
+            self.size.cmp(&other.size)
+        }
+    }
+}
+
+impl PartialOrd for DisplayNode {
+    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
+        Some(self.cmp(other))
+    }
+}
+
+impl PartialEq for DisplayNode {
+    fn eq(&self, other: &Self) -> bool {
+        self.name == other.name && self.size == other.size && self.children == other.children
+    }
+}
+
+impl DisplayNode {
+    pub fn num_siblings(&self) -> u64 {
+        self.children.len() as u64
+    }
+
+    pub fn get_children_from_node(&self, is_reversed: bool) -> impl Iterator<Item = DisplayNode> {
+        if is_reversed {
+            let children: Vec<DisplayNode> = self.children.clone().into_iter().rev().collect();
+            children.into_iter()
+        } else {
+            self.children.clone().into_iter()
+        }
+    }
+}