WIP - dmg/exe installable Letta

[norton120]

Please describe the purpose of this pull request.
This PR will add installable artifacts (dmg and exe) to the CI cycle. they can then be presented as "download Letta" links

How to test
In Phases:

1. ✅ embedded pg: install from ./installable_apps/requirements.txt and then run ./installable_apps/startup.py. Visit http://localhost:8283 and create a new agent, verify that agent saves. Start and stop the service, confirm that the agent persisted.
2. ❌ build installable for MacOS: Instructions TBD
3. ❌ build installable for Windows: Instructions TBD
4. ❌ CI release: Instructions TBD

[norton120]

This is via the embedded Postgres

[norton120]

Working with the updated package (0.5.4)
https://www.loom.com/share/2df983e2d58247899bb5923af6e45d94?sid=5066ddbd-4ed3-4464-86f4-17c11b97487f

[norton120]

added a log viewer via native QT https://www.loom.com/share/a02068bbbeb145e1bb42049d417de5ae?sid=0b809b98-32b5-4990-a8f0-504d2407a138

[norton120]

😠 In adjusting cutelog internals for our uses I am realizing that the QT license for OSS is dangerously restrictive for MemGPT/Letta's model. There's too much interpretation left open - so while having a native x platform log viewer is really nice out of the box, my opinion is that the risk is too high. So we'll need to roll our own simple log displayer. This is frustratingly not a solved problem (if it is and I'm missing it please share).

not a giant lift - if we write to rotating log files and then use one of the billion file parsers out there like (https://github.com/allproxy/json-log-viewer I've used before) or just dumping the log stream into a pretty/letta-branded SPA.
still frustrating.

[norton120]

https://www.loom.com/share/60d431e68da545e995a9862d42b8b501?sid=d3f8b64b-e886-4293-b599-b882fe55e694

web-based log viewer now

[norton120]

running python3 setup.py py2app with a clean venv will build a dist that can be directly archived into a dmg image. But running the dmg image faults because there are missing deps. Installing those deps beforehand (pip3 install -r requirements.txt) introduces a recursion issue somewhere via ast - I think this issue is in llama index, just from process of elimination.

the sys.setRecursionLimit hack gets it to build but fails with missing llama_index even when it's been manually installed - so my theory is that llama_index is failing silently with the hack.

from here I need to try and find a workaround for the recursion issue, and then we have a working dmg installable 👍🏻