Fix subprocess shell usage, error handling, retry logic, and dependency pinning in custom skills

[Copilot]

Addresses five code review findings in the Azure Functions deployment and custom skills code.

Changes

• deploy_azure_functions.py: Removed shell=True from both subprocess.run calls (_deploy_functions and _deploy_functions_withslot) — list-form args require shell=False
• Error handling: Replaced stale requests.exceptions.RequestException catch (left over from a removed REST call) with subprocess.CalledProcessError in both deployment functions
• VectorEmbed/__init__.py: Replaced over-broad retry_if_exception with retry_if_exception_type(TRANSIENT_OPENAI_ERRORS) targeting only rate limit/connection/timeout/server errors; updated log message from misleading "Rate Limit Exceeded" to "Transient error encountered"
• requirements.txt: Pinned all 14 dependencies to explicit versions to eliminate supply chain risk from unpinned installs running under managed identity
• Linting: Split TRANSIENT_OPENAI_ERRORS tuple across lines to fix E501 (line > 120 chars)

# Before: retries on everything, misleading log
@retry(retry=retry_if_exception(lambda e: True), ...)
def _generate_embedding(...):
    ...

# After: targeted retry on transient OpenAI errors only
TRANSIENT_OPENAI_ERRORS = (
    openai.RateLimitError, openai.APIConnectionError,
    openai.APITimeoutError, openai.InternalServerError
)

@retry(retry=retry_if_exception_type(TRANSIENT_OPENAI_ERRORS), ...)
def _generate_embedding(...):
    ...

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.