Add optional GMP-backed bignum multiplication

[Peter1169]

Summary

This adds an opt-in --enable-gmp-bignum configure option that uses LibGMP for large bignum multiplication and squaring.

The default build remains unchanged. When the option is enabled, configure checks for gmp.h and -lgmp, defines ___USE_GMP_BIGNUM for the Gambit runtime library build, and large bignum multiplication tries the GMP path before falling back to Gambit's existing algorithms.

The GMP path keeps Gambit's bignum representation unchanged:

• converts Gambit two's-complement bignums into temporary mpz_t values
• uses mpz_mul for multiplication and squaring
• exports back into a normal movable Gambit bignum
• normalizes the result through the existing bignum normalization path

Related to #761.

Validation

Validated on Linux/WSL and in Docker gcc:9.3.0 with LibGMP installed.

• default build: ./configure, make -j6 from-scratch
• default GAMBITLIB_DEFS and _num.o compile lines do not include -D___USE_GMP_BIGNUM
• default targeted number test passed: ../gsi/gsi -:tl,~~bin=../bin,~~lib=../lib,~~include=../include -f ./run-unit-tests.scm unit-tests/03-number/_num.scm
• GMP build: ./configure --enable-gmp-bignum, make -j6 from-scratch
• GMP build detected gmp.h and __gmpz_mul in -lgmp
• GMP generated lib/_num.c contains the ___USE_GMP_BIGNUM guards and mpz_mul calls
• GAMBITLIB_DEFS contains -D___USE_GMP_BIGNUM in makefile, lib/makefile, and lib/module-common.mk
• _num.o compile lines include -D___USE_GMP_BIGNUM
• GMP targeted number test passed: ../gsi/gsi -:tl,~~bin=../bin,~~lib=../lib,~~include=../include -f ./run-unit-tests.scm unit-tests/03-number/_num.scm
• after rebasing on current master: git diff --check origin/master...HEAD

A local bignum multiplication/squaring probe used non-sparse operands and checked result tails modulo 1000000007; the default and GMP builds produced matching tails.

Case	Default real	GMP real	Speedup	Default allocation	GMP allocation
50k-bit multiply, 20 reps	0.019825s	0.000939s	21.1x	8,127,200	254,080
50k-bit square, 20 reps	0.013814s	0.000669s	20.6x	5,503,200	254,080
200k-bit multiply, 5 reps	0.021878s	0.001537s	14.2x	8,117,880	251,360
200k-bit square, 5 reps	0.015129s	0.001080s	14.0x	5,495,840	251,400

Disclosure: I used Codex (GPT-5.5 xHigh) for a final code review pass after human programming.