Avoid retaining parameter tables for invalid content types

[puffball1567]

Summary

• Initialize the Content-Type parameter table only after validating the media type.
• Preserve the existing ValueError behavior for empty and malformed media types.
• Add empty Content-Type coverage to the existing parser tests.

Why

Prologue calls parseContentType("") for requests without a Content-Type header. The parser previously allocated the parameter table before validating the media type and raising ValueError.

In a persistent Prologue server, those allocations remained live during server operation and grew linearly with the request count. Valgrind reported no definitely-lost blocks at process exit, so this is runtime retention rather than an unreachable shutdown leak. It can still exhaust memory in a long-running server.

Massif attributed about 13.27 MiB of a 13.6 MiB live heap after approximately 5,200 requests to the table initialized in parseContentType. With this change, the measured peak was about 0.33 MiB.

In non-Valgrind runs with one concurrent persistent Joubako client:

• Before: about 108 MiB growth after approximately 26,000 requests.
• After: 48-56 KiB growth after approximately 37,000-39,000 requests.

Validation

• Full Testament suite.
• Existing Content-Type and multipart tests.
• Additional valid, malformed, quoted, and multipart input checks.
• ARC and ORC, with threads both enabled and disabled.
• Valgrind Massif and Memcheck.

This addresses a reproducible contributor to the memory growth reported in #305. The original 700-concurrency reproducer may still have additional contributing factors.

[ringabout]

Thank you! The leak is probably related to nim-lang/Nim#26094

[puffball1567]

Thanks, that issue matches the behavior I observed very closely. This change should also provide a small defensive workaround for existing Nim releases while #26100 is being integrated. Thank you for working on the compiler-side fix!