Opt-in link-time lint warnings, e.g. when ld can be ldh, or jp can be jr

These would be useful as opt-in link-time warnings:

- [ ] For every `ld`, check for and suggest using `ldh` instead if possible.

(This would be an improved substitute for the removed feature of `rgbasm -L`, which optimized known-constant `ld` to `ldh`.)

- [ ] For every `jp`, check for and suggest using `jr` instead if possible.

- [ ] For every `jp` and `jr`, check for redundant jumps to the location immediately after the instruction, and suggest `; fallthrough` instead.

- [ ] For every control flow instruction (`call`, `jp`, `jr`, `rst`) targeting a label in ROMX, do a lint equivalent of `assert BANK(@) == 0 || BANK() == 0 || BANK(@) == BANK()`.

(`rst` always targets ROM0, but if we add support for swappable ROM0, possibly via HDFs (#524), then this would be multi-banked.)

Do this only if PC is in a ROMX section, which also excludes `LOAD`. (In theory, `BANK(@) != 0` for ROMX, but `-t` maps ROMX to ROM0. The destination is then guaranteed to be in ROM0 as well, but that may not remain the case if we add some similar feature to HDFs, so a cheap stringent check for `BANK(@) == 0` is good for future-proofing.)

(Possibly make this a multi-level lint, where by default cross-SECTION jumps are not checked?)

Start by locating the link-time diagnostics and control-flow analysis for ld, jp, jr, call, and rst; the issue names no source files or tests. Define the opt-in and possible multi-level lint behavior, then add checks for ldh/jr substitutions, redundant fallthrough jumps, and ROMX bank targets, with tests showing each warning and its suppression or non-warning cases.

Written by the indexing model from the issue text.