mTLS client prototype

[End-rey]

Continue #4003.

[End-rey]

Benchmark tests were run for the client mTLS variant against the inter-node mTLS. Throughput in ops/s. Small-object GET still benefits in some cases, but overall client-side mTLS looks neutral-to-slightly-negative, while large-object GET generally a bit slower.

REP3

Test	Size, KB	Threads	mTLS	mTLS client	Δ
PUT	4	1	241	243	+1%
PUT	4	4	751	738	−2%
PUT	4	16	1186	1162	−2%
PUT	4096	1	88	88	0%
PUT	4096	4	170	161	−5%
PUT	4096	16	202	187	−7%
GET	4	1	2217	2726	+23%
GET	4	4	4373	4983	+14%
GET	4	16	6142	7027	+14%
GET	4096	1	345	322	−7%
GET	4096	4	575	476	−17%
GET	4096	16	669	528	−21%

EC6+2

Test	Size, KB	Threads	mTLS	mTLS client	Δ
PUT	4	1	167	156	−7%
PUT	4	4	369	322	−13%
PUT	4	16	316	252	−20%
PUT	4096	1	92	92	0%
PUT	4096	4	160	155	−3%
PUT	4096	16	166	163	−2%
GET	4	1	1655	1890	+14%
GET	4	4	3246	3474	+7%
GET	4	16	2301 (8646 err)	1721 (8935 err)	−25%
GET	4096	1	353	326	−8%
GET	4096	4	492	429	−13%
GET	4096	16	547 (509 err)	467 (344 err)	−15%

The main gain vs base still comes from the inter-node changes; adding client-side mTLS does not show a general throughput improvement. The only clear positive signal here is small-object GET at low-to-medium concurrency, while large-object GET is somewhat worse