feat: Advanced Meta Search with KQL Syntax

[pescn]

Summary

Implement advanced search functionality for request metadata using KQL (Kibana Query Language) style syntax, with aggregation support for A/B testing.

Requirements

Query Syntax (KQL Style)

# Basic filtering
model: "gpt-4" AND status: "completed"

# Range queries
duration >= 1000 AND promptTokens < 5000

# Wildcard search
prompt: *error*

# Nested fields (JSONB)
extraHeaders.x-experiment: "group_a"
toolCalls.name: "get_weather"

# Combined queries
(model: "gpt-4" OR model: "claude-3") AND status: "failed"

Aggregation Functions

model: "gpt-4" | stats avg(duration), count(), p95(ttft) by status

Supported functions:

• count() - Count
• avg(field) - Average
• sum(field) - Sum
• min(field), max(field) - Min/Max
• p50(field), p95(field), p99(field) - Percentiles

A/B Testing Support

Users pass experiment tags via extra_header, then filter by that field for comparison.

UI Design (Reference: Kibana Discover)

• Top: Query input with field autocomplete + time range picker
• Middle: Histogram with field breakdown
• Bottom: Document list + field statistics + pagination
• Export: CSV / JSON

Technical Implementation

• Backend: KQL parser → PostgreSQL JSONB queries
• Frontend: Input box with field name autocomplete
• Parser library: chevrotain or custom PEG grammar

---

概要

实现基于 KQL（Kibana 查询语言）风格语法的高级元信息搜索功能，支持聚合函数用于 A/B 测试。

需求

查询语法（KQL 风格）

# 基础过滤
model: "gpt-4" AND status: "completed"

# 范围查询
duration >= 1000 AND promptTokens < 5000

# 通配符搜索
prompt: *error*

# 嵌套字段 (JSONB)
extraHeaders.x-experiment: "group_a"
toolCalls.name: "get_weather"

# 组合查询
(model: "gpt-4" OR model: "claude-3") AND status: "failed"

聚合函数

model: "gpt-4" | stats avg(duration), count(), p95(ttft) by status

支持的函数：

• count() - 计数
• avg(field) - 平均值
• sum(field) - 求和
• min(field), max(field) - 最值
• p50(field), p95(field), p99(field) - 百分位

A/B 测试支持

用户通过 extra_header 传入实验标记，查询时按该字段过滤对比。

界面设计（参考 Kibana Discover）

• 顶部：查询输入框 + 字段自动补全 + 时间范围选择器
• 中间：直方图（可按字段 Breakdown）
• 底部：文档列表 + 字段统计 + 分页
• 导出：CSV / JSON

技术实现

• 后端：KQL 解析器 → PostgreSQL JSONB 查询
• 前端：输入框 + 字段名自动补全
• 解析器库：chevrotain 或自定义 PEG 语法