codegen: force continuation on JAL/JALR resume-PC mismatch instead of aborting caller

[jlsandri]

Stacked on #115 (codegen: handle continuation PCs and nested callee yields). Please merge #115 first — once it lands, this PR's diff shrinks to just the two commits below.

Problem

After a JAL/JALR call + continuation re-entry loop, if ctx->pc didn't match the expected fallthrough address, the previous codegen logged [PC_MISMATCH] and aborted the caller with return.

That abort cascaded up the entire call stack whenever a nested callee legitimately left ctx->pc at an internal resume target — common with overlay functions, nested JALR chains, and any function that sets ctx->pc as part of its own resume-PC mechanism. The caller's stack frame and register context were still valid; the codegen was throwing them away on a false signal.

Fix

Replace the hard return with:

1. A throttled [JALR_FIXUP] / [JAL_FIXUP] diagnostic — 5 warnings per call site, so the signal isn't lost.
2. ctx->pc = fallthroughPc; — force continuation at the caller's expected resume point.

Two commits:

1. JALR resume-PC fixup at the single indirect-call site (depends on codegen: handle continuation PCs and nested callee yields #115 for the re-entry loop structure it edits).
2. JAL resume-PC fixup extending the same treatment to all three direct-call codegen paths for consistency:
   • Direct JAL via findFunctionByAddress (called 0x%x pattern)
   • Direct JAL via lookupFunction (called 0x{literal} pattern)
   • Reloc JAL (called reloc pattern)

Why this is stacked on #115

The JALR hunk rewrites the PC_MISMATCH branch that #115 introduces as part of the re-entry loop. Without #115 merged first, there is no PC_MISMATCH to rewrite.

Scope

• One file: ps2xRecomp/src/lib/code_generator.cpp
• +21 / -7

Rationale

The old behaviour was conservative — any resume-PC anomaly aborted the whole chain — and that's reasonable as a first line of defence. But in practice the only trigger is legitimate internal resume-PC use by the callee, which the caller should tolerate. Demoting the abort to a throttled warning + forced continuation keeps the diagnostic signal while eliminating the cascade-failure class entirely.

[jlsandri]

Closing as part of a batch cleanup after #107 landed. The runtime ecosystem refactor in #107 substantially reworked the files this PR touched, and I would like to re-audit the underlying fix against the new code structure before putting it back in front of you. If the fix is still needed after that re-audit, I will re-open as a focused PR rebased onto current main. Thanks for your patience.