Emit inlinable forms of the FP gate and paired-single helpers

[dougchansan]

Profiling a recompiled Mario Kart: Double Dash on Apple Silicon (sample, race scene) put two runtime helpers near the top of self time inside the dispatch loop:

symbol	self samples	share
ppc_fp_available	869	~7.1%
psq_load_value + psq_store_value	1101	~9.0%

Both are called out of line once per generated instruction site — 121,874 FP sites and 19,642 paired-single sites on this title. ppc_fp_available exists to test one MSR bit that a running game always has set, so the cost was the call rather than the work.

What changed

The emitter now emits ppc_fp_available_inline(), ppc_psq_load_inline() and ppc_psq_store_inline(), so a hosting runtime can provide fast paths. This repo's own runtime gets a real inline gate for the FP case and plain forwarders for paired-single, since it backs the tests rather than a shipping port.

ppc_fp_available() is deliberately still defined — the LLVM backend emits calls to it by name (llvm_runtime_lowering.cpp), so it cannot become header-only.

Measured

With GXRuntime providing the fast paths, alternating A/B with a reversed-order block on an idle host:

change	delta	ranges
inline FP gate	+11.8%	do not overlap
inline paired-single	+4.0%	do not overlap

Both beat their profiled share, because an out-of-line call is also an optimisation barrier: at every generated site the helper arguments (w, gqr_index, indexed, cia) are literals, so inlining lets the compiler fold the LSQE test and the w branch away per site.

Ordering

Requires the hosting runtime to provide the _inline names. Without them a port fails to compile, so this should land after the corresponding RecompCore change (ExpansionPak/RecompCore#9).

codegen_compile asserts generated FP code carries an MSR gate; updated to match the new spelling rather than relaxed.