flint_mpn_divisible

After the trivial cases and the 2-adic part ($b = 2^v B^k b'$ with $b'$ odd), one-limb $b'$ use the Hensel remainder mpn_modexact_1_odd. When both the divisor and the quotient have at least 16 limbs, the operands are first screened modulo $2^{48} - 1$ (which has nine small prime factors; flint_mpn_mod_2exp48m1): a prime dividing $b$ but not $a$ is revealed for about 60% of random pairs. Then, with $m = \mathit{an} - \mathit{bn} + 1$: $b'$ of at most FLINT_MPN_DIVEXACT_SMALL_BN limbs use _flint_mpn_divisible_small; huge operands ($\min(m, \mathit{bn}) \ge$ FLINT_MPN_DIVEXACT_NEWTON_CUTOFF, or $\mathit{bn} \ge$ FLINT_MPN_DIVEXACT_UNBALANCED_CUTOFF and $m \ge 8 \mathit{bn}$) take the candidate quotient from the Newton Hensel division and form only the high part of its product with $b$ (_flint_mpn_mulhigh_known_low), the low part being known; otherwise, as in GMP's mpn_divisible_p, a Hensel division with remainder (ports of mpn_sbpi1_bdiv_qr and mpn_dcpi1_bdiv_qr, the latter from FLINT_MPN_DC_BDIV_QR_CUTOFF limbs) is done after padding $a$ so that its top limb is below that of $b'$, and $b'$ divides $a$ if and only if the remainder equals $b'$.