Meld consecutive template string literals

[alisdair]

When processing a template string, the lexer can emit multiple string literal tokens for what ought to be a single string literal. This occurs when the string contains escape sequences, or consecutive characters which are indistinguishable from escape sequences at tokenization time.

This leads to a confusing AST and causes heuristics about template expressions to fail. Specifically, when parsing a traversal with an index, a key value containing an escape symbol will cause the parser to generate an index expression instead of a traversal.

This commit adds a post-processing step to the template parser to meld any sequences of string literals into a single string literal. Existing tests covered the previous misbehaviour (several of which had comments apologizing for it), and have been updated accordingly.

The new behaviour of the IsStringLiteral method of TemplateExpr is covered with a new set of tests.

---

This change fixes a downstream bug in Terraform which is not currently tracked on GitHub. Apply the following configuration:

resource "null_resource" "none" {
}

Then update the configuration as follows:

resource "null_resource" "none" {
  for_each = toset(["$foo"])
}

moved {
  from = null_resource.none
  to   = null_resource.none["$foo"]
}

This should plan the move operation. Without this change, it results in the following error:

╷
│ Error: Invalid expression
│
│   on main.tf line 7, in moved:
│    7:   to   = null_resource.none["$foo"]
│
│ A single static variable reference is required: only attribute access and
│ indexing with constant keys. No calculations, function calls, template
│ expressions, etc are allowed here.
╵