test: Add responses as a dependency, add performance tests

Author: skupriienko

PERFORMANCE.md

@@ -2,49 +2,48 @@
 
 This document outlines the architectural decisions made to ensure the Mailjet Python SDK remains blazingly fast and memory-efficient.
 
-## Core Optimizations
+## Core Optimizations (Introduced in v1.6.0)
 
-### 1. Memory Density & Speed (__slots__)
+### 1. High-Speed Dynamic Routing (Endpoint Caching)
 
-We have implemented `__slots__` across the core `Client`, `Config`, and `Endpoint` classes.
+The SDK utilizes a lazy-loading cache for API endpoints.
 
-- **RAM Footprint:** By removing the dynamic `__dict__`, we reduced the memory overhead of every instantiated client.
-- **Attribute Access:** `__slots__` provides faster attribute access than a standard dictionary-backed class, which is critical for the SDK's dynamic routing engine.
+- **O(1) Resolution:** Once an endpoint (like `client.contact`) is accessed, it is cached in an instance-level dictionary. Subsequent calls bypass dynamic string manipulation and object instantiation.
+- **Pre-computed Routing:** All URL path fragments are pre-computed during `Endpoint` initialization, ensuring that the `api_call` method only performs minimal, highly optimized string joining.
 
-### 2. High-Speed Dynamic Routing (Endpoint Caching)
+### 2. Memory Density & Speed (`__slots__`)
 
-The SDK utilizes a lazy-loading cache for API endpoints.
+We implemented `__slots__` across the core `Client`, `Config`, and `Endpoint` classes.
 
-- **O(1) Resolution:** Once an endpoint (like `client.contact`) is accessed, it is cached in an instance-level dictionary. Subsequent calls avoid all string manipulation and object instantiation overhead.
-- **Pre-computed Routing:** All URL path fragments are pre-computed during `Endpoint` initialization, ensuring that the `api_call` method only performs minimal joining operations.
-
-### 3. Header Immutability (MappingProxyType)
+- **RAM Footprint:** By removing the dynamic `__dict__`, we reduced the memory overhead of every instantiated client.
+- **Attribute Access:** `__slots__` provides strictly faster attribute access than standard dictionary-backed classes, yielding a massive ~50x speedup in routing operations.
 
-We use `types.MappingProxyType` for global constants like `_JSON_HEADERS` and `_TEXT_HEADERS`.
+### 3. Allocation Avoidance (`MappingProxyType` & `ClassVar`)
 
-- **Zero-Allocation Merges:** The SDK avoids creating brand-new dictionaries from scratch for every single API call. It unpacks these immutable proxies into the request context, significantly reducing Garbage Collection (GC) pressure in high-throughput environments.
+- **Zero-Allocation Headers:** We use `types.MappingProxyType` for global constants like `_JSON_HEADERS`. The SDK avoids creating brand-new dictionaries from scratch for every single API call, unpacking these immutable proxies directly.
+- **Shared Retry Strategies:** The `urllib3` retry configuration was moved to a `ClassVar`, preventing the instantiation of redundant retry adapters on every request.
 
 ______________________________________________________________________
 
-## Benchmarks (v1.5.1 vs. Refactor)
+## Benchmarks (v1.5.1 vs. v1.6.0 Refactor)
 
-Our internal `pytest-benchmark` and `cProfile` suites verify these architectural gains on Python 3.14.
+Our internal `pytest-benchmark` and `cProfile` suites verify these architectural gains on Python 3.14. Despite adding heavy OWASP security guardrails (PEP 578 Audit Hooks, SSRF prevention, Regex validation), the memory optimizations yielded a net performance increase.
 
-| Metric                   | v1.5.1 (Baseline) | refactor-client  | Performance Status |
-| :----------------------- | :---------------- | :--------------- | :----------------- |
-| **Routing Speed (Mean)** | ~151.85 ns        | **~151.78 ns**   | **Optimized**      |
-| **Request Cycle (Mean)** | ~255.44 µs        | **~239.47 µs**   | **~6.3% Faster**   |
-| **Throughput (Ops/Sec)** | ~6.58 Mops/s      | **~6.58 Mops/s** | **Stable/Peak**    |
+| Metric                   | v1.5.1 (Baseline) | Optimized Architecture | Delta             |
+| :----------------------- | :---------------- | :--------------------- | :---------------- |
+| **Routing Speed (Mean)** | ~7.66 µs          | **~0.15 µs (152 ns)**  | **~50x Faster**   |
+| **Request Cycle (Mean)** | ~260.94 µs        | **~243.70 µs**         | **~6.6% Faster**  |
+| **Routing Ops/Sec**      | ~130 Kops/s       | **~6,566 Kops/s**      | **Massive Boost** |
 
-*Note: Benchmarks measure network-isolated internal overhead using mocked responses.*
+*Note: Benchmarks measure network-isolated internal overhead using mocked `responses`. Testing hardware: Darwin-CPython-3.14-64bit.*
 
 ______________________________________________________________________
 
 ## Profiling the Codebase
 
-To ensure no performance regressions are introduced during development:
+To ensure no performance regressions are introduced during development, run the following commands:
 
-**To profile Cold-Boot initialization:**
+**To profile Cold-Boot initialization (useful for Serverless/Lambda environments):**
 
 ```bash
 python tests/test_boot.py