Python from 22x CPython's time to about 2x

Direct calls, fused Python-only instructions, native helpers, a JIT tier for hot loops, per-VM caches and small integers in the value word take the Python benchmark suite from 22.3x CPython 3.13 to about 2.1x.

The starting point, measured on the tools/python_bench.py suite against CPython 3.13: 22.3× its time. f9347730 took it to 8.55× with a new calling convention, per-class inline caches and ten fused Python-only instructions executed out of line so the JavaScript interpreter loop is unchanged. 771eba6b added seventeen more and native json, string, heapq and bisect helpers, to 5.3×. cff586d2 gave the command line a JIT tier for hot Python loops; every fused instruction it does not inline calls the interpreter's own step and continues where that step chose, so results are identical. d6bf5cf0, 6ff55c9f and 2b5e2028 made exceptions, generators and lookups cheaper, to 3.6× with the JIT.

fd18aaf7 was the big step. Python's int is a BigInt in the runtime, so integer code allocated on every operation. A BigInt within ±2^46 is now an immediate in the value word, reusing the null tag's bit pattern with payload bit 47 set, so no new tag was needed and every existing is-a-double check stayed valid. The geometric mean with the JIT fell from 3.6× to 2.1×, and range_loop, int_arith, float_arith, tuple_swap and global_read beat CPython. JavaScript BigInt code got 2-4× faster, other JavaScript is unchanged, and Test262 passes as before.