Bring the Microsoft Teams conference plugin to Zoom parity

[Jamyn]

Is your feature request related to a problem? Please describe.

The Microsoft Teams conference plugin (dispatch_microsoft_teams/conference/) is
substantially behind the Zoom plugin it parallels. Verified against main
(0fe79bda):

	Teams	Zoom
Methods	create	create, delete, add_participant, remove_participant
Metrics (counter/timer)	silently absent	working
Config doc	none	configuring-zoom.mdx
Tests	none	none (parity)

The metrics gap is a real defect rather than an omission. apply (src/dispatch/decorators.py:164)
is a class decorator — it iterates cls.__dict__ and rewrites each callable
attribute. Teams applies it to the create method (plugin.py:25-26), whose
__dict__ is empty, so the loop body never runs and the function is returned
untouched. Zoom applies it to the class (plugin.py:53-54) and is instrumented.
No crash, no warning — just missing telemetry.

Additional defects in dispatch_microsoft_teams/conference/client.py:

• logger.info(f"Masked Result is {result}") (line 53) logs the entire MSAL
  token response, including access_token. Nothing is masked despite the label.
  Mitigating: the default LOG_LEVEL resolves to ERROR, so it only fires when
  an operator raises the level to INFO/DEBUG — i.e. exactly while debugging.
• requests.post (line 65) has no timeout and can hang indefinitely.
• No HTTP status check. A Graph error body flows into meeting_info["joinWebUrl"]
  (plugin.py:38) → KeyError → swallowed by the broad except → create
  returns None.
• recordAutomatically and isPasscodeRequired are sent as JSON strings
  ("true"/"false") rather than booleans. Inferred — not tested against live
  Graph.
• That broad except also means the real exception never reaches
  conference/flows.py, so the incident event log records a generic "plugin
  encountered an error" (flows.py:39) instead of the cause. Zoom lets it
  propagate into the except at flows.py:28, which logs the reason.

Missing delete is not a live leak — nothing in the codebase calls
conference.delete; Zoom's is dead code too. Worth noting so it is not
prioritized as a resource leak.

Describe the solution you'd like

Bring Teams to Zoom parity:

1. Move @apply(counter, ...)/@apply(timer, ...) from the method to the class
   so metrics actually emit.
2. Add a timeout to the Graph call and check the HTTP status; raise on error
   rather than returning a body that fails downstream with KeyError.
3. Stop logging the MSAL result, or log only the non-secret fields.
4. Narrow or remove the broad except in create so the real reason reaches the
   incident event log.
5. Send recordAutomatically/isPasscodeRequired as booleans.
6. Add configuring-microsoft-teams.mdx alongside configuring-zoom.mdx.
7. Implement delete/add_participant/remove_participant for interface parity
   (low priority — unused by callers).

Describe alternatives you've considered

Fixing only the metrics decorator. Rejected: the token logging and the missing
timeout are the changes with actual operator impact, and they sit in the same two
files.

Guarding against the apply-on-a-method misuse inside apply itself (raise when
handed a non-class). Reasonable as a follow-up so the mistake cannot recur
silently, but it is a separate change from making Teams work.

Additional context

Line references are against 0fe79bda. Every claim above is verified against
source except the boolean-vs-string Graph payload, which is inferred — it was
not tested against live Graph and needs a real tenant to confirm.