fix[issue #142]: SIGSEV during concurrent rendering of PDF in Image() or ImageDPI()

[intx4]

Summary

This PR fixes intermittent SIGSEGVs during concurrent rendering by adding proper MuPDF lock wiring during context creation and tightening resource lifetimes in the cgo path.

Root cause: fz_new_context_imp was called with nil lock callbacks, which violates MuPDF’s multithreading contract when rendering overlaps across contexts/goroutines.

What changed

• Added MuPDF lock callbacks (fz_locks_context) in cgo:

  • Implemented lock/unlock callbacks and FZ_LOCK_MAX mutex array.
  • Initialized once per process and shared by all contexts.
  • Passed lock context into fz_new_context_imp(...) for every Document.

• Hardened context/document lifecycle:

  • Centralized context creation in shared init path.
  • Added constructor cleanup on partial failures.
  • Made Close() cleanup ordering explicit and defensive (stream -> doc -> context -> C buffer).

• Fixed NewFromMemory lifetime hazards:

  • Switched to C-owned byte buffer (C.CBytes) instead of relying on Go slice memory.
  • Kept stream reference with fz_keep_stream and released temporary stream ref.
  • Freed C buffer on Close().

• Added/updated documentation:

  • README now clearly states:
    • parallel rendering is supported across separate Document instances,
    • concurrency on the same Document (including racing with Close) is not supported.
  • Added code comments in fitz_cgo.go documenting same-document concurrency limits and lock-free method expectations.

Why this fixes the crash

MuPDF requires caller-provided locks for safe multithreaded use. With lock callbacks registered, internal shared paths (alloc/glyph/freetype and related global state) are synchronized correctly across concurrent rendering workloads, preventing unsafe overlap that could lead to SIGSEGV.

Compatibility

• Existing call sites (New, NewFromMemory, NewFromReader, ImageDPI, etc.) continue to work unchanged.

Tests

• Added a test file to verify the fix
• You need to provide your own concurrency.pdf file for the test to work. I could not share the one I used as it contains sensitive data (it is an artifact from my app production environment)

Notes

• This PR intentionally does not add same-document concurrent access support.
• Callers should continue using one Document per worker/goroutine for parallel conversion workflows.

[intx4]

This fixes #142