* Replacing `.point.` with `.`

* Adding exceptions

As recommended by include-what-you-use, partly replaced by forward-declarations

* sed -i 's/\([^a-zA-Z0-9_]\)points.size/\1size/g'
* Replace all casts for `width = EXPR size()`
* Format string changes
* Auto
* Range based for loop
* OpenMP doesn't like range for loop

* Reuse code, add exception in cuda

Added release of lock, else a deadlock could happen, though probably only happening when debugging.

* Remove unneeded`#include <gtest/gtest.h>`

* Replace ‘gtest/gtest.h’ with ‘pcl/test/gtest.h’

* Include ‘gtest/gtest.h’ instead of itself

Co-authored-by: NAaryaman Vasishta @jammm <jem456.vasishta@gmail.com>
Shrijit Singh @shrijitsingh99 <shrijitsingh99@gmail.com>

Use using in test class, and reuse standard countWithinDistance implementation in SIMD implementations

The function doSamplesVerifyModel is called with a percentage of points (previously: 10 percent, which is about 330 points)
and checks if ALL these points are within the threshold. Since the plane in the test data set has about 2500 inliers out of
total 3283 points (about 75 percent), the probability that even 10 randomly chosen points ALL belong to the plane is just
0.75^10, about 5 percent. The probability for 330 points is practically zero. I had a look at the paper this implementation
is based on. If I understand the experiments correctly, they make that test with just 1 or 2 points. So a better percentage
for the unit test would be 0.1 percent (about 3 points).

Fix issues reported by run-clang-tidy -header-filter='.*' -checks='-*,bugprone-too-small-loop-variable'

All input points are on the ground plane. This test passes after #2767
and fails before.

* commands are now always in lowercase
* remove unnecessary expressions in else/endif
* unified whitespaces between command and bracket

The test compiles differntly if the configuration is Release or Debug. In Debug mode, threads are required to finish their task in 15 seconds instead of 1.

Closes 2585

Revise direction test in SampleConsensusModelParallelLine/RANSAC to be consistent with set tolerance.

- Include directories are now added for opt dependencies
- Make module dependencies propagate to the corresponding tests

Additionaly rename test_sample_consensus_line_model to reflect the fact
that now it tests multiple line models.

Avoid re-computation of normals every time the test is run and remove
dependency on features and kdtree modules.

Google test macros a supposed to have expected value as the first
parameter and actual value as the second.