Reorganize mrb_cache_entry and mrb_method_t types

[dearblue]

The purpose is to remove the mid field from the mrb_cache_entry structure. The resulting RAM requirement for the method cache is reduced from 5 words per entry to 4 words per entry for 32-bit CPUs.

The relevant changes are as follows:

• Removed MRB_USE_METHOD_T_STRUCT.

  The mrb_method_t type is now always defined as a structure.

• Include method IDs in mrb_method_t

  Change the flags member to uint32_t. The bitstring structure should be the same as the keys of the mt table in class.c.

I believe the impact on API compatibility with previous versions is minimal.

---

Attached is a benchmark comparison for your reference.
The targets are (1) the current master, (2) this PR, (3) another alternative and (4) 3.3.0.
Another alternative is to move the bits embedded in struct mrb_cache_entry::mid to struct mrb_cache_entry::c0 (like "word-boxing"). Link to difference.

[dearblue]

The following is the number of CPU cycles obtained from valgrind --tool=cachegrind -- bin/mruby benchmark/bm_fib.rb.
The order of the lines is the same as in the previous benchmark.
The first group is the results from gcc13 -O3 and the second group is the results from gcc13 -Os.

40,186,751,990 (100.0%)  PROGRAM TOTALS         master
40,656,510,074 (100.0%)  PROGRAM TOTALS         this PR
40,733,987,325 (100.0%)  PROGRAM TOTALS         another alternative
40,577,677,399 (100.0%)  PROGRAM TOTALS         3.3.0

61,020,950,744 (100.0%)  PROGRAM TOTALS         master
61,021,647,478 (100.0%)  PROGRAM TOTALS         this PR
61,177,299,954 (100.0%)  PROGRAM TOTALS         another alternative
61,333,670,456 (100.0%)  PROGRAM TOTALS         3.3.0

# I only recently learned about the role and usage of Cachegrind from the SQLite documentation.