Remove the nested read loops in FastCGIConn, hide complexity inside StreamBuffer

buffer.h

@@ -8,23 +8,19 @@ class ConstBuffer {
 	ConstBuffer(const char *buf, size_t len);
 
 	[[nodiscard]] size_t ReadMaxLen() const;
-	[[nodiscard]] const char *Read(size_t len);
+	[[nodiscard]] virtual const char *Read(size_t len);
 	template<class T> [[nodiscard]] const T *ReadObj();
 
 	bool Discard(size_t len); // like Read() but don't use the result
-	void ResetRead(); // next read from last commit
-	void Commit(); // commit read position
 
   protected:
 	const char *const_buf_;
 	size_t len_;
 	size_t start_ = 0;
-	size_t commit_ = 0;
 };
 
 class Buffer : public ConstBuffer {
   public:
-  	Buffer(const char *buf, size_t len) = delete;
 	Buffer(char *buf, size_t size, size_t len);
   	Buffer(size_t size);
 
@@ -32,9 +28,9 @@ class Buffer : public ConstBuffer {
 	[[nodiscard]] size_t WriteMaxLen() const;
 	void Wrote(size_t len);
 
-	void Consume(); // discard up to last commit
+	void Commit(); // commit read position
 
-  private:
+  protected:
   	std::unique_ptr<char> own_buf_;
 	char *buf_;
 	const size_t size_;