wip: add CBOR fast path decode

[ongyimeng]

Summary

Felt.UnmarshalCBOR previously used fxamacker/cbor’s reflection-based decoder. Since MarshalCBOR always emits a 4-element array of uint64 limbs, decode that shape directly and fall back to the generic decoder for other valid encodings.

Benchmark

Baseline using generic CBOR decode vs fast path, -count=8, benchstat:

sec/op improvement:

Benchmark	Baseline	Fast path	Δ
StateUpdateByBlockNum p50	71.1µs	38.0µs	−46.5%
StateUpdateByBlockNum p95	558µs	265µs	−52.5%
StateUpdateByBlockNum p99	555µs	264µs	−52.5%
BlockByNumber p50	620µs	376µs	−39.4%
BlockByNumber p95	3.53ms	1.70ms	−51.9%
BlockByNumber p99	5.07ms	2.39ms	−52.8%

All results are highly significant: p=0.000, n=8.

B/op and allocs/op are unchanged at approximately 0%, as expected. The fast path avoids reflection CPU overhead in cbor.Unmarshal, while decoded felts allocate the same way.

[codecov]

Codecov Report

❌ Patch coverage is 82.14286% with 10 lines in your changes missing coverage. Please review.
✅ Project coverage is 75.25%. Comparing base (c68a109) to head (68dfa67).

Files with missing lines	Patch %	Lines
core/felt/felt.go	82.14%	6 Missing and 4 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3792      +/-   ##
==========================================
+ Coverage   75.16%   75.25%   +0.08%     
==========================================
  Files         438      438              
  Lines       39577    39633      +56     
==========================================
+ Hits        29748    29824      +76     
+ Misses       7735     7720      -15     
+ Partials     2094     2089       -5     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.