Limit log length for `SchemaError`

[stakahashy]

Is your feature request related to a problem? Please describe.
When SchemaError occurs in pandera, it outputs the full log of all schema-errored elements. This becomes problematic when validating large dataframes with many validation failures, as the error log can become tremendous - potentially thousands of lines long. This makes it difficult to read error messages, can crash terminals or notebooks （this is the main motivation why I raised this issue and implemented), and makes debugging harder as the important information gets buried in excessive output.

Describe the solution you'd like
Add a configurable limit to the number of failure cases shown in SchemaError messages. The implementation should:

• Set a default limit of 100 failure cases shown
• Allow configuration via environment variable PANDERA_MAX_FAILURE_CASES
• Support runtime configuration through config_context(max_failure_cases=n)
• Use -1 as a special value to show all failure cases (unlimited)
• Use 0 to show only the count without examples
• When truncated, display a message like "... and X more failure cases (Y total)"

Describe alternatives you've considered

1. Fixed hard limit: Simply truncate at a fixed number (e.g., always show 10) - rejected because different use cases need different limits
2. Log to file: Dump full errors to a file instead of console - rejected as it adds complexity and doesn't solve the immediate visibility problem
3. Sampling: Show a random sample of errors - rejected because the first N errors are often more informative than a random sample
4. Using None for unlimited: Considered using Optional[int] with None meaning unlimited - rejected in favor of -1 as sentinel value to keep the type as int

Additional context
#2095

This feature is particularly important for:

• Data pipelines processing large datasets (millions of rows)
• Initial data exploration where many validation rules might fail
• Production environments where excessive logging can impact performance
• Jupyter notebooks where huge outputs can make the notebook unresponsive

Example of current behavior with a 1M row dataframe having 500K failures:

  SchemaError: Column 'amount' failed validator: greater_than(0) failure cases: -1.5, -2.3, -0.1, -5.2, ... [500,000 values printed]

With this feature (default limit=100):

  SchemaError: Column 'amount' failed validator: greater_than(0) failure cases: -1.5, -2.3, -0.1, -5.2, ... [96 more values] ... and 499,900 more failure cases (500,000 total)