fix #28990, improve top-level location info

[JeffBezanson]

This fixes #28990, plus the following issue:
Before:

julia> error()
ERROR: 
Stacktrace:
 [1] error() at ./error.jl:42
 [2] top-level scope at none:0

After:

julia> error()
ERROR: 
Stacktrace:
 [1] error() at ./error.jl:42
 [2] top-level scope at REPL[1]:1

This works by (1) passing an initial filename and line number to lowering (so that locations set in a :toplevel expression can apply to the following expression), and (2) making the REPL use the same parsing function that files (and include_string) use, which puts a line node before each expression.

Base.parse_input_line now wraps everything in a :toplevel expression including a line node.

The low-level parser entry points are simplified and renamed to parse-one, parse-all, and parse-file.

The ex === () check in parse I believe was dead code; it looks like something that was left out when we replaced () with nothing a long time ago. All parsing functions return nothing for parse(""), so giving an error here would be breaking but we can consider whether we want to switch to that.

Question 1: I believe the code in interpreter.c setting jl_lineno (which this PR removes) is not necessary, since we should now be able to get line numbers from the interpreter stack trace mechanism. Can anyone think of a reason we need that code?

Question 2: I don't understand the need for the change to repl_filename. Without it, the filename for the first repl input is REPL[0] if the input is very simple, e.g. error() as in the example above. But if I insert even just a single space (within the same input cell) before error(), then you get REPL[1].

[vtjnash]

Can anyone think of a reason we need that code?

I find it can be invaluable for debugging segfaults, when the stack can be all smashed up and such. It's usually the first thing I type when opening gdb. Doesn't need to be particularly reliable (try/finally) or even exported for that use-case though.

[vtjnash]

I don't think the parser has any deprecations, so this support code is dead and should be removed

[JeffBezanson]

We might add more parser deprecations in the future.