Fix TaskParameterTaskItem serialization perf

[ccastanedaucf]

Fixes

Fixes several hotspots I found while profiling RAR service serialization.

Changes Made

• Fix effectively doubling the work in the write direction, where TaskParameterTaskItem is created to wrap an existing TaskItem, but then of its metadata is parsed one again right before serialization
• Use MSBuildNameIgnoreCaseComparer.Default (by moving it into Framework`
• Avoid additional call to retrieve value when enumerating dictionary returned by ITaskItem2.CloneCustomMetadataEscaped(). If this is a Utilities.TaskItem coming back from a task, doing an accessor check is non-negligible overhead due to CopyOnWriteDictionary.
• SImplify translation logic. Prefer unidirectional ITranslator APIs over manually checking each direction or reimplementing methods, and let TaskParameterTaskItem handle its own serialization.

A bunch of small improvements but they add up. Since the out of proc TaskHost is the only piece currently using this, here's some profiles off my RAR service builds.

TaskParameterTaskItem.CopyMetadataTo()

This is often hit by ReferenceTable.SetItemMetadata in RAR. This fix here was to use IMetadataContainer.ImportMetadata() bulk set, since ImmutableDictionary performs better when operations are batched.

Before:

After:

Double ITaskItem parsing

Essentially when in the write direction the old code would construct a TaskParameterTaskItem, but later cast it to ITaskItem and parse everything out again. You can see this here where CloneCustomMetadataEscaped() is hit even though by this point we've already extracted the external TaskItem to our own instance.

Before:

After:

[ccastanedaucf]

lol probably broke something while pulling this out of my RAR branch, will check UT failures later today

[ccastanedaucf]

Updated with profile traces now that I've fully isolated it from other perf stuff. Tests passing so should be good to go 👍

[SimaTian]

Quite large change to parse out so I'm not 100% sure I didn't miss something, however overall I like it.
It get's rid of a bunch of code, which is nice.
It has a performance benefit which is even nicer.
It passes tests, which is good.

Since it is only being used by TaskHost(and eventually RAR caching), I'm voting for.

[SimaTian]

Perf DDRITs have passed:
https://dev.azure.com/devdiv/DevDiv/_git/VS/pullrequest/646407
There is some SDK error in pipelines but from a first glance it appears to be unrelated. Can you check it though please?
https://devdiv.visualstudio.com/DevDiv/_build/results?buildId=11829685&view=logs&j=3095faff-5e5f-54e0-ff01-49e86a1b9dd7&t=db3cb6fb-4173-593d-fda5-ee991ff8c0ad

[ccastanedaucf]

FYI I removed the unescaped item spec struct since I expect memory footprint per TaskItem to be more important now w/ the task host, and there's ways to fix that one in RAR. Everything else is good!

[YuliiaKovalova]

some of the failures from the experimental run might be relevant, asked in DM to verify.