[Optimizer] Optimize variable-length edge predicate execution

[czpmango]

Introduction
Cypher provides predicate functions to filter all variable-length edges, and the statement is like:

MATCH (v:player)-[e:like*2..5]->(n)
WHERE all(i in e where i.likeness > 90)
RETURN n

There is room for optimization in the current implementation of nebula, and we can consider embedding the filter expression into the traverse operator, optimized plan is like this:
Scan -> Traverse(allEdgeFilter_ = e.likeness>90) -> Project

Contents
Add related optimization rules to gain performance benefits.

Related work
Some query semantics that are difficult to express in go statements can be replaced by match statement.

[wey-gu]

It's exciting to see we started to optimize towards functions that used to be always evaluated after data fetching!

[Sophie-Xie]

#5464
#5470
#5481
#5503

[czpmango]

Only the basic edge all predicate pushdown is supported(predicate is a first-level attribute expression), such as:

MATCH (v)-[e*2..5]-(n) where all(i in e where i.prop1 >3 and i.prop2<4) RETURN *

Unsupported list:

1. function predicate like

MATCH (v)-[e*2..5]-(n) where all(i in e where properties(i).prop1 >3 and rank(i).prop2<4) RETURN *

2. double-not expression

MATCH (v)-[e*2..5]-(n) where all(i in e where !!(i.prop1 >3) and i.prop2<4) and not not all(i in e where i.prop>4) RETURN *

3. ...