chore(ci): simplify import profiler metrics and remove P90/P99 stats

[hebaalazzeh]

This PR improves the import profiler tool ci check (#17657) by simplifying its output metrics:

• Simplified & Reliable Stats: Shows only Min, P50 (Median), Max, and StdDev. Removed the calculation and formatting of P90, P99, and Mean stats as they are not reliable for the standard 10-iteration sample size.
• Enhanced Readability: Reordered the output values from lowest to highest (Min -> P50 -> Max).
• Sample Size Indicator: Included the number of iterations ([N=...]) in the title of each metric section to make sample sizes explicit.
• Cleaned Up Dead Code: Removed the unused _calculate_percentiles helper function to keep the codebase clean.

[gemini-code-assist]

Code Review

This pull request introduces an import_profile test step in the CI pipeline and enhances the Python SDK import profiler to support package-based module auto-detection, cache burn-in iteration discarding, and performance regression checks against baseline CSVs. The reviewer feedback highlights several key improvements for robustness and testability, including quoting shell variables in the CI script, preventing potential TypeError and StopIteration exceptions during CSV parsing and parameter handling, and refactoring run_master to return exit codes instead of calling sys.exit() directly.