[onert] In-place Reshape and ExpandDims operation

[hseok-oh]

We are always using memcpy for Reshape and ExpandDims operation.
If these operation's input is used only once and updated shape does not require padding, we can skip copy operation.

For this, we need to modify memory allocation and tensor handling implementation.

[mbencer]

Due to recommendation from #12814 (comment) I'm assigning myself to this issue ;)

[mbencer]

@hseok-oh Could you please take a look at #14057?
I am not sure if you plan implementation similar to my draft or maybe more global approach (like creating common buffer during tensor creation). The second one approach has much larger affected area.

[mbencer]

@hseok-oh Handling of sharing memory for static shapes is already on master. Should dynamic shape support be processed also under this issue or is it better to create a separate issue for it?
BTW.: Should I treat dynamic shapes part as something with big priority?

[hseok-oh]

@hseok-oh Handling of sharing memory for static shapes is already on master. Should dynamic shape support be processed also under this issue or is it better to create a separate issue for it?

It doesn't matter either way.

BTW.: Should I treat dynamic shapes part as something with big priority?

Treating dynamic shape is more useful. But considering its impact on performance, It may not has a very high priority.