Integrate internal changes

[shelkesagar29]

This PR integrates the following internal changes into OSS.

commit c6f9d4b57ae793b08ddf415529a83e2aca6df5ce
Author: Sagar Shelke [email protected]
Date: Tue Jun 10 14:27:27 2025 -0700

[Dialect/Plan] Add constant foldable subgraph analysis pass.

This MR adds a generic pass for constant foldable subgraph analysis pass. This pass
implements the following three things,
1. A forward data flow analysis to find constant foldable ops.
2. Use clustering to find constant foldable subgraphs.
3. Outline constant foldable subgraphs to `func.func` with
`plan.constant_foldable` attributes.
Pass option `skipClustering` allows user to extend default pass behavior
and skip certain ops from outlining.

MLIR test cases are added including region ops tests.

commit 5fad6480597ee15bdb4a8c04c033b7ef6beff2d0
Author: Christopher Bate [email protected]
Date: Tue Jun 10 19:27:04 2025 +0000

NFC: remove unused functions

commit 7eb853f52525756e399a7fc6fdc4ab25244d6f94
Author: Chris Bate [email protected]
Date: Tue Jun 10 12:06:51 2025 -0700

[compiler] Create InferTensorValueRangeInterface

Previously, our TensorValueBoundsAnalysis was only able to analyze values
that were either constant or were produced by the result of `plan.with_values`.
This worked because we rely on `plan.with_values` to link integer range
analysis and the TensorValueBoundsAnalysis. However, after clustering there
can occur edge cases where the `plan.with_values` op is not present.
Furthermore, eventually we would like to avoid creating `plan.with_values`
ops in the first place since they temporarily bloat the IR.

This commit adds a new interface, InferTensorValueRangeInterface, that allows
operations to participate in the TensorValueBoundsAnalysis. The logic
for `plan.with_values` is implemented using the interface, and this change
adds an implementation and test for `stablehlo.convert`. Besides this
new behavior for `stablehlo.convert`, this change is a pure refactor and
no other new behavior is added.

commit d49fe69e08d79db07bf9f801427c97f6bfbdfcda
Author: Christopher Bate [email protected]
Date: Fri Jun 6 17:03:45 2025 +0000

NFC: Move `python` under `integrations/python`

Moves the `python` directory under `integrations/python` to better
reflect that, like PJRT, the Python packages are a downstream consumer of
the MLIR-TensorRT compiler/runtime libraries.

commit ef5490368b2235e80bb21a6aca39db43d24f1c47
Author: Sagar Shelke [email protected]
Date: Sat Jun 7 14:27:21 2025 -0700

[tensorrt] Implement inliner inferface for TensorRT dialect

This MR implements `DialectInlinerInterface` for the TensorRT dialect.
All ops in TensorRT dialect are pure and thus can be inlined, except `module`,
`call` and `call_alloc`. Upstream `--inline` pass doesn't inline operation with
  callop interface (`call` and `call_alloc`) because these operations in TensorRT dialect
are referring to callee in different symbol table comapred to their own.

MLIR valid and invalid test is added.

commit cd56aa6a511e2091fcd86106f20d27ff3673db75
Author: Christopher Bate [email protected]
Date: Wed Jun 11 14:26:03 2025 +000

Fix build with BUILD_SHARED_LIBS=ON

The new InferTensorValueRangeInterface was used without correctly
specifying the library dependency the PlanIR and StablehloExtIR
libraries.

GitOrigin-RevId: cd56aa6a511e2091fcd86106f20d27ff3673db75