[Hexagon] Optimize post-increment load and stores in loops.

[sgundapa]

This patch optimizes the post-increment instructions so that we can packetize them together.
v1 = phi(v0, v3')
v2,v3 = post_load v1, 4
v2',v3'= post_load v3, 4

This can be optimized in two ways

v1 = phi(v0, v3')
v2,v3' = post_load v1, 8
v2' = load v1, 4

[steven-johnson]

This has broken HVX Codegen in Halide; our tests now fail with

vsub(v*:*.h,v*:*.h)  Failed Opcode is : 2656 (V6_vL32b_pred_ai)
No offset range is defined for this opcode. Please define it in the above switch statement!
UNREACHABLE executed at /home/halidenightly/build_bot/worker/llvm-19-x86-64-linux/llvm-project/llvm/lib/Target/Hexagon/HexagonInstrInfo.cpp:2981!

(see e.g. https://buildbot.halide-lang.org/master/#/builders/118/builds/86/steps/12/logs/correctness_simd_op_check_hvx)

Please either roll back this change ASAP or offer a fix-forward as this is blocking Halide from passing our tests.

[sgundapa]

I will take a look and fix it.

[sgundapa]

I am currently working on a fix

[sgundapa]

Fix undr review here: #82386