Shift responsibility to stringify message content from Loader to runtime layer

[vladsud]

This (in future) will help us avoid double strigification for scenarios like chunked ops, compressed ops.

[msfluid-bot]

⯅ @fluid-example/bundle-size-tests: +364 Bytes

Metric Name	Baseline Size	Compare Size	Size Diff
aqueduct.js	391.54 KB	391.64 KB	⯅ +100 Bytes
connectionState.js	680 Bytes	680 Bytes	■ No change
containerRuntime.js	190.91 KB	191.01 KB	⯅ +100 Bytes
loader.js	151.74 KB	151.9 KB	⯅ +164 Bytes
map.js	42.63 KB	42.63 KB	■ No change
matrix.js	131.29 KB	131.29 KB	■ No change
odspDriver.js	150.24 KB	150.24 KB	■ No change
odspPrefetchSnapshot.js	38.4 KB	38.4 KB	■ No change
sharedString.js	152.53 KB	152.53 KB	■ No change
Total Size	1.25 MB	1.25 MB	⯅ +364 Bytes

---

Baseline commit: dcf9599

Generated by 🚫 dangerJS against d10ffc7

[anthony-murphy]

can we do the same thing for signals? It will keep the layers consistent across our message types

[anthony-murphy]

slightly unrelated, but we also process signals in the delta manger, specifically we json parse them, and loose data, like the ref seq, which prevent us for leveraging ref seq on signals in the runtime. maybe this should be a parallel work stream to get the treatment between the two symmetrical, and ideally basically transparent to the loader layer.

[vladsud]

I agree, but do not want to make these PRs more complex.
https://dev.azure.com/fluidframework/internal/_workitems/edit/1841

[github-actions]

This commit is queued for merging with the next branch! Please ignore this PR for now. Contact @microsoft/fluid-cr-infra for help.

[markfields]

@vladsud - I don't think we can make any changes like this (breaking so-called forwardCompat or backCompat) for IContainerContext in main. In this case, an old implementation does not reflect the new interface. I know you add a runtime check, but this is actually causing a type compatibility problem in Bohemia codebase. I think we can work around it but just bringing to your attention.

[vladsud]

I think you are right. I assumed that interface would be used only internally, but it's clearly wrong assumption.