check-templates: Avoid duplicate tokenizing step.

Now we only tokenize the file once, and we pass
**validated** tokens to the pretty printer.

There are a few reasons for this:

    * It obviously saves a lot of extra computation
      just in terms of tokenization.

    * It allows our validator to add fields
      to the Token objects that help the pretty
      printer.

I also removed/tweaked a lot of legacy tests for
pretty_print.py that were exercising bizarrely
formatted HTML that we now simply ban during the
validation phase.

tools/lib/template_parser.py

@@ -316,7 +316,7 @@ def tag_flavor(token: Token) -> Optional[str]:
         raise AssertionError(f"tools programmer neglected to handle {kind} tokens")
 
 
-def validate(fn: Optional[str] = None, text: Optional[str] = None) -> None:
+def validate(fn: Optional[str] = None, text: Optional[str] = None) -> List[Token]:
     assert fn or text
 
     if fn is None:
@@ -445,6 +445,8 @@ def validate(fn: Optional[str] = None, text: Optional[str] = None) -> None:
 
     ensure_matching_indentation(fn, tokens, lines)
 
+    return tokens
+
 
 def ensure_matching_indentation(fn: str, tokens: List[Token], lines: List[str]) -> None:
     for token in tokens: