Become more robust to parse errors

I've started working on a sql [lsp server implementation](https://microsoft.github.io/language-server-protocol/specifications/lsp/3.17/specification/), using `sqlparser` as the parser. Unfortunately, I very quickly hit a wall, mostly around how `sqlparser` discovers and surfaces errors. For instance, let's say the user types the following: `SELECT * F` and hits the key. A good lsp server would suggest completing `F` to `FROM`, even just do it automatically in that case. If one tries to parse that input string with `sqlparser`, it returns a `ParserError("Expected end of statement, found: F")`. The error isn't unexpected, but now (a) since the ast isn't available in the face of errors, there are no recommendations/commands/lints I can make based on a error-prone AST, and (b) I don't have enough information (line/col info) to send a diagnostic (the thing that makes the red squiggly lines) to the client. The line/col information was suggested in #179, but a fair amount of the work that would enable a really good LSP server implementation is being able to work with an AST that has errors in it.

I _suspect_ what I'm describing is _really_ hard, and has all sorts of backwards-incompatible changes that would be required to make this work. I think it'd be worthwhile, though.

Start by reading the sqlparser ParserError handling and the parser entry points involved when parsing `SELECT * F`; the issue provides no specific files or tests. Compare the current behavior with the LSP diagnostic requirements and determine how an error-tolerant AST, completion support, and line/column diagnostics should be defined. Done should include agreed behavior and coverage for incomplete SQL such as this example.

Written by the indexing model from the issue text.