Allow fatal SBCL errors to be caught

[kartik-s]

Problem

A call into Lisp may trigger a fatal error in SBCL, dropping the process unceremoniously into LDB and making it difficult for the user to gracefully terminate the process.

Solution

Use setjmp to save the execution context before calling into Lisp, and use longjmp to return to the pre-call execution context with a special error code if a fatal SBCL error was encountered.

Notes

• This feature requires patching SBCL to make the lossage_handler function pointer non-static, setting sbcl-librarian::*non-static-lossage-handler* to T, and adding a fatal-error code to the error map.
• DEFINE-ERROR-TYPE takes a new fatal-error argument specifying which value to return when a Lisp call triggers a fatal SBCL error.
• The names of the function pointers in the generated C bindings are now prefixed with an underscore.
• Each function pointer is wrapped by a function definition with the original callable name. The body of each function definition performs a setjmp before calling into Lisp through the associated function pointer.
• If there is no error-map or no fatal-code is specified, then the fatal error handler just drops into LDB.

[macrologist]

Just a first pass. I didn't try to understand what it is doing.