address some concerns in initial device_update PR

[bmc-msft]

This PR adds the following changes:

• moves to enabling the client to support light-weight clones rather than managing lifetimes for the client.
• removes the need for the client to be mutable
• moves to using azure_identity's AutoRefreshingTokenCredential rather than by-hand token refresh.
• replaces runtime panics with returning errors
• moves helper functions to return the DeserializeOwned trait, rather than strings which get immediately deserialized
• moves many of the errors to use the azure_core error types (in prep for moving to the pipeline archtecture)
• moves to using query_pairs_mut to add URL params, rather than building query strings by hand
• simplfies creating the mock client used for testing

Things still to do (in follow-up PRs):

• Move to using azure_svc_deviceupdate under the hood
• Rather than iteratively fetching all the results at once before returning to the client, move to using a streaming api, akin to azure_data_tables::requests::QueryEntityBuilder.stream

[rylev]

This looks good. I think we should maybe start moving over to the pipeline architecture ASAP as we might be doing some work that will just go away once the transition is made. I, however, still own a written transition guide...