feat: make max query topk configurable

[Rm1n90]

Summary

• Adds a max_query_topk parameter to zvec.init() so users can raise the hard topk ceiling (default 1024) when they accept the extra memory/latency tradeoff
• Stores the configured limit in GlobalConfig and reads it in VectorQuery::validate() instead of using the hardcoded kMaxQueryTopk constant
• Exposes the same config through the C API via zvec_config_data_{set,get}_max_query_topk()

Usage

import zvec

zvec.init(max_query_topk=4096)
results = collection.query(vq, topk=2048)  # now works

zvec_config_data_t *cfg = zvec_config_data_create();
zvec_config_data_set_max_query_topk(cfg, 4096);
zvec_initialize(cfg);

Files changed (7)

Layer	File	Change
C++ core	src/include/zvec/db/config.h	Add max_query_topk to ConfigData + accessor
C++ core	src/db/common/config.cc	Default value (1024) + validation
C++ core	src/db/index/common/doc.cc	Use configurable limit in VectorQuery::validate()
Python	python/zvec/zvec.py	New max_query_topk param with type/range validation
Python binding	src/binding/python/model/common/python_config.cc	Parse and validate from config dict
C API	src/include/zvec/c_api.h	Declare getter/setter
C API	src/binding/c/c_api.cc	Implement getter/setter

Test plan

• Full build from source (macOS ARM64, Python 3.13)
• Python test suite: 1009 passed, 54 skipped, 0 failures
• Validated edge cases: float, negative, zero, bool, string all correctly rejected with clear error messages
• Default behavior unchanged — limit stays 1024 unless explicitly overridden

🤖 Generated with Claude Code

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Your free trial has ended. If you'd like to continue receiving code reviews, you can add a payment method here.