Performance improvements

[mkrd]

Benchmark                                 Baseline    Current     Speedup
─────────────────────────────────────────────────────────────────────────
BDouble converging to two for 3000 steps  382ms       378ms       1.01x
10^300_000                                351ms       111ms       3.16x
Factorial of 40_000                       313ms       88ms        3.56x
Fib 100000                                280ms       4ms         70.00x
Matrix ^ 500                              448ms       440ms       1.02x
Mersennes to 2^512                        262ms       263ms       1.00x
Get 15000! as String                      206ms       67ms        3.07x
Factorial 30_000 as String                185ms       184ms       1.01x
Perm and Comb (8000, 4000)                207ms       183ms       1.13x
Multiply 270k x 270k bit                  169ms       26ms        6.50x
Multiply 70M x 1k bit                     162ms       161ms       1.01x
Divide 200k by 100k bit                   418ms       421ms       0.99x
Divide 10M by 1k bit                      479ms       486ms       0.99x
GCD of two 130k-bit BInts                 516ms       514ms       1.00x
mod_exp 1800-bit                          438ms       440ms       1.00x
BDouble division 100k-bit                 310ms       306ms       1.01x
BDouble decimal expansion 70k digits      344ms       110ms       3.13x
1500! to hex string                       368ms       3ms         122.67x
Shift 1M-bit left by 5M, 100 times        353ms       347ms       1.02x