fix: unsafe error handling and allocation crashes

[hamchapman]

These changes address multiple critical safety issues:

Issue #118 - Large allocation crash:

• Replace .map() with progressive array building in readN()
• Prevents memory allocation crashes from malformed CBOR with huge declared lengths
• Add capacity reservation capped at reasonable size (1024)

Issue #69 - Array bounds checking:

• Add bounds validation before accessing data ranges in UnkeyedDecodingContainer
• Prevents crashes from corrupted data with out-of-bounds ranges
• Throw proper DecodingError instead of crashing

Unsafe try! replacements:

• AnyCodingKey: Replace try! with proper error handling for invalid key types
• KeyedDecodingContainer: Replace try! with proper error propagation
• KeyedDecodingContainer: Replace fatalError() with throwing DecodingError
• UnkeyedDecodingContainer: Remove fatalError(), allow errors to propagate
• AnyCodingKey.encode(): Replace fatalError() with throwing EncodingError
• AnyCodingKey.key(): Replace fatalError() with preconditionFailure (better messages)

Encoder changes:

• Make encodeMap() and encodeMapChunk() throw instead of using try!
• Update call sites with proper error handling
• Add preconditionFailure for programming errors (forbidNonStringMapKeys violations)

---

Fixes: #118
Fixes: #69