Omit unnecessary field from location range

[pocke]

This PR reduces memory usage for RBS::Location by omitting unnecessary fields.

Previously RBS::Location had byte pos, char pos, line, and column. But they are caches except for char pos. These caches are rarely accessed. So we can drop them from RBS::Location.

design doc: https://hackmd.io/QNkMU6roTqSn_iXb2ENhsA

Benchmarking

I used benchmark/memory_new_rails_env.rb for the benchmark.
It decreases 15% memory.

# Before
Total retained:  52372539 bytes (459882 objects)

allocated memory by class
-----------------------------------
 152307312  Hash
  20146528  RBS::Location
  15744181  String
  15124312  Array


# After
Total retained:  44234355 bytes (459882 objects)

allocated memory by class
-----------------------------------
 152307312  Hash
  15744181  String
  15124312  Array
  12008296  RBS::Location

---

This PR also fixes two problems of the current implementations.

First, Buffer#pos_to_loc returns the wrong line and column for the EOS position of a string without the end of a newline. For example, the location for abc| (| is the cursor) should be [1, 3], but #pos_to_loc unexpectedly returns [2, 0].
This problem is fixed in 81b44e1

Second, #pos_to_loc returns an unexpected value when the position is located EOF.
The cause is that the parser determines the length of EOF as 1. Then, the RBS::Location of the pEOF token contains an out-of-range position. Because an EOF has a length in C String, but it doesn't have a length in Ruby String.
I changed the length of EOF token to zero to fix this problem.
16006c2

[soutaro]

Great work! 👏

RBS::Location now stores rbs_loc_range, a pair of start char pos and end char pos, but other C internal code keeps using range, with additional fields. Because the range is internal and not exposed as Ruby objects, this PR reduces the memory usage after parsing. 👍

[soutaro]

🎉