Experimental python updates: memory, embeddings, native functions, chat w/ memory

[jjhenkel]

Motivation and Context

In an effort to port more of the original code over to Python, this PR adds the following:

1. Lots of little fixes to the code (as I've read through, double-checked, and ran some more examples)
2. The addition of our first core_skill in Python --- TextMemorySkill (inferring the delegate type is tricky here without C#'s nice reflection capabilities... probably need some way to "error" if a Skill's functions aren't fully type-annotated as that's how I figured out the right delegate type)
3. The addition of decorators to mark up a native function (similar to the attributes in the C# version)
4. A minimal working memory subsystem (the C# code was hard to trace, especially given that I needed to reconcile custom embeddings-code and our numpy-based impl, probably bugs to work out there)
5. An end-to-end example of chatting w/ memory!

Description

At this point, I'd say we have maybe ~60/70% of the core (as it exists right now on the main branch) functionality in Python. We can define semantic functions, template prompts (that have Var/Code blocks), define native functions, work with memory/embeddings, etc. We're missing a lot of tests and things like chunking text. We're also missing everything related to the planner.

There's now some low-hanging fruit here to address for the port:

1. We should start figuring out tests and moving many tests over
2. We should add AzureOpenAI support (mostly just a dupe of openai support, will use the same calls to the openai package and everything, just needs to override certain params)
3. We should add in the rest of the missing core skills
4. We should look at loading skills from directories (haven't tried this yet, probably missing some basic things here)
5. We should port more examples! (I have a "basic" one, a "chat" one, and now a "chat with memory")

Contribution Checklist

N/A for now, just trying to get all of the code ported.

[alexchaomander]

Overall LGTM. I think it'd be great to get this into the experimental branch so that we have a fuller picture of the Python Semantic Kernel with embeddings and memory.