[Swift] Reduce swift_beginAccess cost to optimize performance

[chaokunyang]

Feature Request

Reduce swift_beginAccess cost to optimize performance

Is your feature request related to a problem? Please describe

Baseline Benchmark Numbers (commit In chaokunyang@cd360e2)

Benchmark	ns/op	ops/sec
Struct Serialize	58.60	17,064,577
Struct Deserialize	66.56	15,024,390
Sample Serialize	300.69	3,325,648
Sample Deserialize	942.51	1,061,000
MediaContent Serialize	686.30	1,457,097
MediaContent Deserialize	1669.27	599,065
StructList Serialize	364.29	2,745,061
StructList Deserialize	406.82	2,458,083
SampleList Serialize	1420.34	704,059
SampleList Deserialize	4986.70	200,533
MediaContentList Serialize	2994.44	333,952
MediaContentList Deserialize	8014.31	124,777

#3457 (HEAD) Performance

Benchmark	ns/op	ops/sec	Regression
Struct Serialize	106.13	9,422,352	1.81x slower
Struct Deserialize	87.18	11,470,519	1.31x slower
Sample Serialize	290.24	3,445,375	OK (slightly faster)
Sample Deserialize	995.39	1,004,627	1.06x slower
MediaContent Serialize	688.33	1,452,791	OK
MediaContent Deserialize	1635.96	611,262	OK (slightly faster)
StructList Serialize	366.24	2,730,416	OK
StructList Deserialize	376.84	2,653,632	OK (faster)
SampleList Serialize	1445.54	691,784	OK
SampleList Deserialize	4992.00	200,320	OK
MediaContentList Serialize	2991.15	334,320	OK
MediaContentList Deserialize	CRASH	CRASH	FATAL ERROR

Key Issues

1. Struct Serialize: 1.81x regression (58.60 -> 106.13 ns/op)
2. Struct Deserialize: 1.31x regression (66.56 -> 87.18 ns/op)

The cost are mostly in swift Access checks

Describe the solution you'd like

Find an approach to reduce such costs

Describe alternatives you've considered

#3355
#3457

Additional context

No response