Modified Explain to be able to print spliced (optimized) subqueries within the Explain out.
Unfortunately I had to add an additional variable to the Subquery Start Node, this has potential to be dumped again later. The variable has no "Production" semantics.

Jenkins:
http://jenkins01.arangodb.biz:8080/view/PR/job/arangodb-matrix-pr/6816/

Non-spliced subquery:

Query String (94 chars, cacheable: true):
 FOR x IN 1..2 LET y = (FOR z IN 1..2 LET a = (FOR b IN 1..2 RETURN b) RETURN [x,z,a]) RETURN y

Execution plan:
 Id   NodeType            Est.   Comment
  1   SingletonNode          1   * ROOT
  2   CalculationNode        1     - LET #9 = 1 .. 2   /* range */   /* simple expression */
  3   EnumerateListNode      2     - FOR x IN #9   /* list iteration */
 14   SubqueryNode           2       - LET y = ...   /* subquery */
  4   SingletonNode          1         * ROOT
  5   CalculationNode        1           - LET #11 = 1 .. 2   /* range */   /* simple expression */
  6   EnumerateListNode      2           - FOR z IN #11   /* list iteration */
 11   SubqueryNode           2             - LET a = ...   /* const subquery */
  7   SingletonNode          1               * ROOT
  8   CalculationNode        1                 - LET #13 = 1 .. 2   /* range */   /* simple expression */
  9   EnumerateListNode      2                 - FOR b IN #13   /* list iteration */
 10   ReturnNode             2                   - RETURN b
 12   CalculationNode        2             - LET #15 = [ x, z, a ]   /* simple expression */
 13   ReturnNode             2             - RETURN #15
 15   ReturnNode             2       - RETURN y

Indexes used:
 none

Optimization rules applied:
 none

Spliced subquery:

Query String (94 chars, cacheable: true):
 FOR x IN 1..2 LET y = (FOR z IN 1..2 LET a = (FOR b IN 1..2 RETURN b) RETURN [x,z,a]) RETURN y

Execution plan:
 Id   NodeType            Est.   Comment
  1   SingletonNode          1   * ROOT
  2   CalculationNode        1     - LET #9 = 1 .. 2   /* range */   /* simple expression */
  3   EnumerateListNode      2     - FOR x IN #9   /* list iteration */
 16   SubqueryStartNode      2       - LET y = ( /* subquery begin */
  5   CalculationNode        2         - LET #11 = 1 .. 2   /* range */   /* simple expression */
  6   EnumerateListNode      4         - FOR z IN #11   /* list iteration */
 18   SubqueryStartNode      4           - LET a = ( /* subquery begin */
  8   CalculationNode        4             - LET #13 = 1 .. 2   /* range */   /* simple expression */
  9   EnumerateListNode      8             - FOR b IN #13   /* list iteration */
 10   ReturnNode             8               - RETURN b
 19   SubqueryEndNode        8           - ) /* subquery end */
 12   CalculationNode        8           - LET #15 = [ x, z, a ]   /* simple expression */
 13   ReturnNode             8           - RETURN #15
 17   SubqueryEndNode        8       - ) /* subquery end */
 15   ReturnNode             8       - RETURN y

Indexes used:
 none

Optimization rules applied:
 Id   RuleName
  1   splice-subqueries

NOTE:
The estimated cost is different.
The non-optimized code prints cost within subqueries only for one round trip, and exposes only the total mount of executed subqueries to the following nodes
The optimized code prints cost assumptions accumulated for all subquery executions.

Need discussion if we want to adapt this or not.