Audit: Review ROUTING_FPS_PER_PAGE 4096, and multiple uses of hard-coded 32 in routing_filter.c

This issue was identified during code-walk thru of PR #230. There are these references in routing_filter.c, which need a better understanding, commenting in the file.

24 #define ROUTING_FPS_PER_PAGE 4096

During discussion, it appeared that 4096 is some sort of "upper-bound" on number of routing filters, but it wasn't clear whether this should be related in any way to page size [etc.]

 399       MATRIX_ROWS * MATRIX_COLS + // matrix
 400       ROUTING_FPS_PER_PAGE +      // fp_buffer
 401       ROUTING_FPS_PER_PAGE +      // old_fp_buffer
 402       ROUTING_FPS_PER_PAGE / 32;  // encoding_buffer

 413    old_fp_buffer = fp_buffer + ROUTING_FPS_PER_PAGE;
 414    encoding_buffer = (uint64 *)(old_fp_buffer + ROUTING_FPS_PER_PAGE);
 415    memset(encoding_buffer, 0xff, ROUTING_FPS_PER_PAGE / 32 * sizeof(uint32));

There are many more instances of hard-coded '32' ... and it's not clear what the math is all about here. E.g.

 147 routing_get_index(uint32 fp,
 148                   size_t index_remainder_and_value_size)
 149 {
 150    return index_remainder_and_value_size == 32
 151       ? 0 : fp >> index_remainder_and_value_size;

 239       // ffs returns the index + 1 ALEX: I think that's what we want though.
 240       bit_offset = __builtin_ffs(encoding_word) - 1;
 241       *end = 32 * word + bit_offset;

 446    for (uint32 new_fp_no = 0; new_fp_no < num_new_fp; new_fp_no++) {
 447       new_fp_arr[new_fp_no] >>= 32 - cfg->fingerprint_size;
 448       new_fp_arr[new_fp_no] <<= value_size;

  698       platform_assert(cfg->fingerprint_size + value_size <= 32);

 811    uint32 fp = hash(slice_data(key), slice_length(key), seed);
 812    fp >>= 32 - cfg->fingerprint_size;

And we should find a way to give a semantic #define for 32, and use that consistently where applicable in this file.