Perf: support sending JDWP commands asynchronously

[testforstephen]

This PR provides some experimental performance tuning for remote debugging:

• Add a thread cache to reduce JDWP frequency for thread info. threads request is the most frequent DAP request, and adding a cache can save lots of unnecessary JDWP queries.
• Provide async mode to process JDWP commands in parallel. This hopes to improve the DAP responsiveness on high-latency networks.

[testforstephen]

Here is the DAP response speed comparison when using Java debugger to attach debug a remote application.

Application: spring-petclinic (https://github.com/spring-projects/spring-petclinic)
Total threads on breakpoint: 20
Total stacks on breakpoint: 98
Local variables: 5

Sample 1: Base network latency for a JDWP command is 95ms

DAP Request	Response (ms) on @0.43.0	@next (without async)	@next (async)
initialize	4	1	0
attach	547	591	592
setBreakpoints	1906	1863	295 (6X)
setExceptionBreakpoints	7	1	0
configurationDone	194	194	251
threads	4349	2208 (2X)	471 (9X)
threads	4311	96 (45X)	96 (45X)
stackTrace	8943	6380 (1.4X)	1781 (5X)
scopes	4	0	0
variables	9914	11795	674 (15X)
next	2685	770 (3X)	577 (4.5X)
threads	4320	106 (40X)	95 (45X)
stackTrace	387	194 (2X)	192 (2X)
scopes	1	0	0
variables	1551	1545	389 (4X)
continue	2710	194 (14X)	190 (14X)
disconnect	99	98	96

Sample 2: Base network latency for a JDWP command is 284 ms

DAP Request	Response (ms) on @0.43.0	@next (without async)	@next (async)
initialize	3	0	3
attach	1619	1726	1911
setBreakpoints	5571	5450	870 (6X)
setExceptionBreakpoints	4	0	6
configurationDone	576	568	1103
threads	13090	6565 (2X)	1475 (9X)
threads	12979	282 (46X)	288 (46X)
stackTrace	26620	18744 (1.4X)	5355 (5X)
scopes	6	0	3
variables	29134	34947	2035 (14X)
next	7488	2280 (3X)	1730 (4X)
threads	12922	286 (45X)	287 (45X)
stackTrace	1152	568 (2X)	576 (2X)
scopes	0	0	0
variables	4604	4543	1150 (4X)
continue	8038	559 (14X)	579 (14X)
disconnect	288	288	287