feature: support custom headers for LLM call when using custom embedding

[planetf1]

Adds support when specifying custom headers when using CUSTOM_EMBEDDING_MODEL & associated parameters
milvus only (only milvus supports custom embeddings)

Example:

CUSTOM_EMBEDDING_HEADERS="MY_LLM_SECRET=0123456789abcdef"

or

CUSTOM_EMBEDDING_HEADERS='{"MY_LLM_SECRET": "0123456789abcdef"}'

Alternatively multiple values can be comma separated, or the variable can be specified in JSON

Tested on a platform which requires custom headers for authentication

Fixes #61

[planetf1]

I had to add 4 type corrections in the milvus test that were added in a previous PR (unrelated to this change) - lint now runs cleanly

[planetf1]

Parsing of environment in start.sh was also fixed to be more robust with quoted values

[ajbozarth]

Tested local using RITS and this LGTM.

One big caveat: You need to remove the cli changes and open a PR on the new repo with those instead since it got moved out after you opened this

[planetf1]

Removed cli changes. Will open issue on new cli repo.
The change didn't actually enable custom embeddings - it was just a command I noticed was missing from the cli, whilst it was present in the mcp server (and a useful command I was using to debug!)

[planetf1]

Raised issue AI4quantum/maestro-cli#3 for cli changes