Better Error Propagation

[cali-jumptrading]

We should propagate errors as structs with a type and error code, and only set the error into the task info context once to avoid error overwrite errors and clean up code.

This would look like returning a struct that looks like this:

struct __attribute__((packed)) fd_exec_result {
  /* kind and err matching agave */
  uchar kind;
  schar err;

  /* custom err is a u32 in agave */
  uint custom_err;
};
typedef struct fd_exec_result fd_exec_result_t;

For VM errors, the proposed returned error struct would look like this:

struct fd_vm_result {
  fd_exec_result_t exec_res;
  int err; // this can be a schar to save space
};
typedef struct fd_vm_result fd_vm_result_t;

Open to other suggestions as well! The motivation for this proposal is to remove the need for the developer to remember to set error details via macros like https://github.com/firedancer-io/firedancer/blob/main/src/flamenco/runtime/context/fd_exec_txn_ctx.h#L169. Using these macros before returning can be easily forgotten or accidentally called more than once up the call stack (when returning errors).

Example of forgotten ERR_FOR_LOG: #4025
Example of overwritten error: https://github.com/firedancer-io/firedancer/pull/4069/files#diff-7068ff1660560ece7e7d6a7c6848146e115fc8ad40dd3addaab552703d70b15eR414