[Coverity] GPU Plugin Medium issues (#27761)

### Details:
- This PR fixes: 1562334, 1559879, 1559863, 1559854, 1559677, 1559879,
1559863

- 1559461 and 1559460 are related to ov::pass::pattern::matcher in
`decompose_reduce_scalar_output.cpp`
- 1559886 and 1559881 are related to ov::pass::pattern::matcher in
`transpose_fusion.cpp`
- 1559862 is related to ov::pass::pattern::matcher in
`transpose_fusion.cpp`
- 1559698 and 1559691 are related to ov::pass::pattern::matcher in
`bcast_and_pad_zp_buffers.cpp`

- all member variables are initialized in the constructors' initializer
lists of `ov::pass::pattern::matcher` to address 1559461, 1559460,
1559886, 1559881, 1559862, 1559698, 1559691.

### Tickets:
 - [*CVS-153064*](https://jira.devtools.intel.com/browse/CVS-153064)
 - CVS-145082

---------

Co-authored-by: Pavel Durandin <[email protected]>

src/core/include/openvino/pass/pattern/matcher.hpp

@@ -62,20 +62,45 @@ class OPENVINO_API Matcher {
     // Avoid implicit string construction from nullptr.
     Matcher(const std::shared_ptr<Node> pattern_node, std::nullptr_t name) = delete;
 
-    Matcher() = default;
-    Matcher(Output<Node>& pattern_node) : m_pattern_node{pattern_node} {}
-
-    Matcher(Output<Node>& pattern_node, const std::string& name) : m_pattern_node(pattern_node), m_name{name} {}
+    Matcher()
+        : m_match_root{},
+          m_pattern_node{},
+          m_pattern_map{},
+          m_pattern_value_maps{},
+          m_matched_list{},
+          m_name{""},
+          m_strict_mode{false} {}
+    Matcher(Output<Node>& pattern_node)
+        : m_match_root{},
+          m_pattern_node{pattern_node},
+          m_pattern_map{},
+          m_pattern_value_maps{},
+          m_matched_list{},
+          m_name{""},
+          m_strict_mode{false} {}
+
+    Matcher(Output<Node>& pattern_node, const std::string& name)
+        : m_match_root{},
+          m_pattern_node{pattern_node},
+          m_pattern_map{},
+          m_pattern_value_maps{},
+          m_matched_list{},
+          m_name{name},
+          m_strict_mode{false} {}
 
     /// \brief Constructs a Matcher object
     ///
     /// \param pattern_node is a pattern sub graph that will be matched against input graphs
     /// \param name is a string which is used for logging and disabling a matcher
     /// \param strict_mode forces a matcher to consider shapes and ET of nodes
     Matcher(const Output<Node>& pattern_node, const std::string& name, bool strict_mode)
-        : m_pattern_node(pattern_node),
-          m_name(name),
-          m_strict_mode(strict_mode) {}
+        : m_match_root{},
+          m_pattern_node{pattern_node},
+          m_pattern_map{},
+          m_pattern_value_maps{},
+          m_matched_list{},
+          m_name{name},
+          m_strict_mode{strict_mode} {}
 
     // Some matches should start on a node rather than an output. These three constructors
     // are transition until we work out the right way to do that.

src/plugins/intel_gpu/include/intel_gpu/primitives/resample.hpp

@@ -15,7 +15,7 @@ namespace cldnn {
 struct resample : public primitive_base<resample> {
     CLDNN_DECLARE_PRIMITIVE(resample)
 
-    resample() : primitive_base("", {}) {}
+    resample() : primitive_base("", {}), scales_port(0) {}
 
     using InterpolateOp = ov::op::util::InterpolateBase;
 

src/plugins/intel_gpu/src/graph/impls/onednn/convolution_onednn.cpp

@@ -121,7 +121,7 @@ struct convolution_onednn : typed_primitive_onednn_impl<convolution> {
 
 private:
     int _zero_point_mask;
-    dnnl::memory::data_type _wzp_data_type;
+    dnnl::memory::data_type _wzp_data_type = dnnl::memory::data_type::undef;
 
 protected:
     std::unique_ptr<primitive_impl> clone() const override {

src/plugins/intel_gpu/src/kernel_selector/kernels/fully_connected/fully_connected_kernel_bf_tiled.cpp

@@ -846,9 +846,11 @@ void FullyConnected_bf_tiled::GetUpdateDispatchDataFunc(KernelData& kd) const {
                         // quantized input is char type
                         kd.internalBufferSizes.push_back(input_size);
                         // half type of de_quan_scale and activation sum for each quantized group
+                        OPENVINO_ASSERT(quantize_grp_size != 0, "Error: quantize_grp_size is zero.");
                         kd.internalBufferSizes.push_back((input_size / quantize_grp_size) * 2 * 2);
                     }
 
+                    OPENVINO_ASSERT(quantize_grp_size != 0, "Error: quantize_grp_size is zero.");
                     kd.kernels[0].params.workGroups.global = {std::max((input_size / quantize_grp_size), (size_t)1), 1, 1};
                     kd.kernels[0].params.workGroups.local = {16, 1, 1};
                 }
@@ -983,6 +985,7 @@ KernelsData FullyConnected_bf_tiled::GetMultiKernelsData(const Params &params,
     const auto& fc_params = static_cast<const fully_connected_params&>(params);
 
     size_t quantize_grp_size = get_dynamic_quantize_group_size(fc_params);
+    OPENVINO_ASSERT(quantize_grp_size != 0, "Error: quantize_grp_size is zero.");
 
     bool bProperInput = fc_params.inputs[0].GetLayout() == dl;
     if (!bProperInput && !fc_params.inputs[0].PitchesDifferFromLogicalDims()) {

src/plugins/intel_gpu/src/plugin/transformations/kv_cache_fusion.cpp

@@ -63,7 +63,7 @@ KVCacheFusionMatcher::KVCacheFusionMatcher() {
             return false;
 
         // TODO: Support conversion internally
-        if (concat_node->get_output_element_type(0) != past_node->get_output_element_type(0))
+        if (!concat_node || concat_node->get_output_element_type(0) != past_node->get_output_element_type(0))
             return false;
 
         auto variable = past_node->get_variable();