improve deserialize error message

[GregPlowman]

Ref: #31337 (comment)

The values were written with the same Serializer context, so there can be shared references within the stream. They need to be deserialized using the same context object as well.

If this is not the case, then an error can be triggered with the somewhat cryptic message:

ERROR: KeyError: key 0 not found

This is an attempt to improve that error message.
Maybe the new message is too verbose or imprecise?
Or maybe there are other conditions which trigger the error?
In any case I submit this for your consideration.

[StefanKarpinski]

Much better. I think it might be clearer to say that if the serialize calls use the same serializer object then the corresponding deserialize calls should also use the same serializer object.

[JeffBezanson]

I would worry about the performance implications of this double lookup --- in the common case the key will be found.

[GregPlowman]

Can we use get(s.table, id, -1)?

Or perhaps get(s.table, id, secret_table_token)?

[GregPlowman]

Updated to use this form:

get(s.table, id) do
    errmsg = ...
    error(errmsg)
end

[GregPlowman]

Wow, CI overload. I'm overwhelmed.
Some passed, some failed. I don't know how to interpret this.
Are any of the errors related to this PR?

[StefanKarpinski]

Yeah, I have to confess the new CI output is pretty confusing. What's going on with this, @staticfloat?

[staticfloat]

Yes, lots of new CI. This new CI has a separate checkmark for Julia building on each platform, and for testing on each platform. So, for instance, you can see that this PR built successfully on linux i686, linux x86_64, linux aarch64, and macOS x86_64. It did not test successfully on any of them though. Clicking details on the right of the tester_linux64 builder, shows that the error was due to CI being broken. I'm going to close and re-open the PR to re-trigger the CI and we'll see if you get more green checkmarks this time around. :)

We're still shaking out all the bugs of this CI system, and some of the problems (on systems like armv7l, which weren't tested very thoroughly before) have required merging some commits into master, so if you want all the green checkmarks, then you'll probably need to rebase this branch on top of master. That being said, for the next couple days, I would look primarily at linux32, linux64, macos64, win32 and win64, as those should be issue-free already.

[GregPlowman]

Closing and reopening to restart CI

[StefanKarpinski]

Travis failures are unrelated.