Migrate bwa-mem2 from zlib to zlib-ng

[brainstorm]

Companion PR to #56242, let's see how much more perf is there with this switch.

[dslarm]

Here's some new data for this - I ran with LD_LIBRARY_PATH pointing to installations of cloudflare zlib (github.com/cloudflare/zlib ) and zlib-ng (github.com/zlib-ng/zlib-ng) - quite easy to install - so we could check this is going to be worth doing:

TL;DR - it's worth it, for all platforms. At 64 threads, it's looking like a 5-10% performance improvement - although no benefit at say 8 or 16 threads.

MIN of time		threads
platform	combo	8	16	32	64
c7a.24xlarge (AMD Genoa)	llvm	196.87	121.24	83.44	84.45
	llvm-cloudflare	198.65	115.93	82.88	78.33
	llvm-zlib-ng	202.4	109.05	84.18	78.33
	orig	220.57	121.82	84.71	85.97
c7a.24xlarge Total		196.87	109.05	82.88	78.33
c7g.16xlarge (AWS Graviton3)	llvm	216.03	122.35	82.93	82.86
	llvm-cloudflare	218.6	119.82	82.06	79.91
	llvm-zlib-ng	218.84	119.04	81.67	78.46
	llvm-zlib-ng-clang19	220.56	118.56	84.97	78.8
	orig	241.2	130.41	87.1	84.78
	zlib-ng	218.39	119.88	82.02	82.94
c7g.16xlarge Total		216.03	118.56	81.67	78.46
c7i.24xlarge-smt-off (Intel Sapphire Rapids, SMT off)	llvm	163.25	95.45	82.14	88.31
	llvm-cloudflare	163.33	95.2	82.4	82.19
	llvm-zlib-ng	163.38	93.96	76.99	82.1
	orig	190.84	105.04	82.53	91.65
c7i.24xlarge-smt-off Total		163.25	93.96	76.99	82.1
c7i.24xlarge-smt-on (Intel Sapphire Rapids, SMT on)	llvm	163.85	94.91	83.9	81.82
	llvm-cloudflare	163.51	93.89	77.08	78.31
	llvm-zlib-ng	162.73	95.35	77.34	78.01
	orig	189.21	106.26	87.55	86.37
c7i.24xlarge-smt-on Total		162.73	93.89	77.08	78.01
c8g.24xlarge (AWS Graviton4)	llvm	182.5	100.56	77.84	78.8
	llvm-cloudflare	188.31	100.47	78.47	83.46
	llvm-cloudflare-clang19	191.96	100.55	78.5	78.32
	llvm-zlib-ng	185.84	100.74	79.19	76.3
	llvm-zlib-ng-clang19	189.28	100.22	75.2	82.92
	orig	200.74	110.87	83.43	78.37
c8g.24xlarge Total		182.5	100.22	75.2	76.3

orig = current gcc build in bioconda, and the rest are your llvm artifacts with the LD_LIBRARY_PATH set to the other compression libraries. FWIW - it's not worth using LLVM for the compression library itself.