Add support for extend primop

Nickel has change some primops to contain a forward slash `/`, which
wasn't supported by the parser (they would be parsed as modulo instead).

This commit extend the `builtin` rule to accepts a larger class of valid
builtin names, including those new ones, and hopefully potential future
ones as well.

corpus/basic.txt

@@ -139,8 +139,7 @@ identifier (builtin function)
       (applicative
         (record_operand
           (atom
-            (builtin
-              (ident))))))))
+            (builtin)))))))
 
 ================================================================================
 number int

corpus/primop.txt

@@ -0,0 +1,59 @@
+================================================================================
+Normal primop
+================================================================================
+
+%plus% 1 2
+
+--------------------------------------------------------------------------------
+
+(term
+  (uni_term
+    (infix_expr
+      (applicative
+        (applicative
+          (applicative
+            (record_operand
+              (atom
+                (builtin))))
+          (record_operand
+            (atom
+              (num_literal))))
+        (record_operand
+          (atom
+            (num_literal)))))))
+
+================================================================================
+Slashed primop
+================================================================================
+
+%array/length% []
+
+--------------------------------------------------------------------------------
+
+(term
+  (uni_term
+    (infix_expr
+      (applicative
+        (applicative
+          (record_operand
+            (atom
+              (builtin))))
+        (record_operand
+          (atom))))))
+
+================================================================================
+Extended primop (currently inexistent)
+================================================================================
+
+%__some/primop.foo.bar'%
+
+--------------------------------------------------------------------------------
+
+(term
+  (uni_term
+    (infix_expr
+      (applicative
+        (record_operand
+          (atom
+            (builtin)))))))
+

grammar.js

@@ -463,9 +463,15 @@ module.exports = grammar({
     //grammar.lalrpop: 509
     // Different from lalrpop grammar, we parse all possible builtins, not just
     // the defined ones.
-    builtin: $ => seq(
+    builtin: _ => seq(
       "%",
-      $.ident,
+      // We are a bit more liberal in what can go in a builtin functions than
+      // for identifiers, because they are properly delimited by `%` and
+      // upstream Nickel added `/` as a valid character already, so there's a
+      // precendent for extensions (although it's not very likely). We just
+      // exclude spaces and make sure the builtin starts with either a latter or
+      // an undescore, to ensure reasonable names.
+      /_*[a-zA-Z][a-zA-Z0-9./_'-]*/,
       "%",
     ),
 

src/grammar.json

@@ -1909,8 +1909,8 @@
           "value": "%"
         },
         {
-          "type": "SYMBOL",
-          "name": "ident"
+          "type": "PATTERN",
+          "value": "_*[a-zA-Z][a-zA-Z0-9./_'-]*"
         },
         {
           "type": "STRING",

src/node-types.json

@@ -239,17 +239,7 @@
   {
     "type": "builtin",
     "named": true,
-    "fields": {},
-    "children": {
-      "multiple": false,
-      "required": true,
-      "types": [
-        {
-          "type": "ident",
-          "named": true
-        }
-      ]
-    }
+    "fields": {}
   },
   {
     "type": "chunk_expr",