Optimize decoding CBOR text string by up to 16%

[fxamacker]

This PR reduces decoding duration for CBOR text string by up to 16% (depending on the data structure and values being decoded) on existing benchmarks.

This PR optimizes decoding of CBOR text string by:

• using fast path for parsing CBOR text strings up to 23 bytes, which covers most map string keys and struct field names
• using fast path to set Go strings
• using fast path for map string keys to bypass the generic decoding path (e.g. map[string]any)

Benchmarks

Benchmark results for Unmarshal() omit allocs/op and B/op because they are unchanged:

goos: linux
goarch: amd64
pkg: github.com/fxamacker/cbor/v2

NOTE: These benchmark results were produced on a PC running other processes.
Please measure on your own hardware with appropriate settings and workloads.

                                                            │     old     │                 new                 │
                                                            │   sec/op    │   sec/op     vs base                │
Unmarshal/CBOR_map_to_Go_interface_{}                        1.303µ ± 0%   1.231µ ± 0%   -5.56% (p=0.000 n=10)
Unmarshal/CBOR_map_to_Go_map[string]interface_{}             1.433µ ± 0%   1.275µ ± 0%  -11.06% (p=0.000 n=10)
Unmarshal/CBOR_map_to_Go_map[string]string                   993.6n ± 0%   832.5n ± 0%  -16.21% (p=0.000 n=10)
Unmarshal/CBOR_indef-len_map_to_Go_interface_{}              1.410µ ± 0%   1.319µ ± 0%   -6.46% (p=0.000 n=10)
Unmarshal/CBOR_indef-len_map_to_Go_map[string]interface_{}   1.770µ ± 0%   1.561µ ± 0%  -11.78% (p=0.000 n=10)
Unmarshal/CBOR_indef-len_map_to_Go_map[string]string         1.288µ ± 0%   1.099µ ± 0%  -14.67% (p=0.000 n=10)
Unmarshal/CBOR_map_to_Go_map[string]any                      2.688µ ± 0%   2.539µ ± 0%   -5.53% (p=0.000 n=10)
Unmarshal/CBOR_map_to_Go_struct                              1.916µ ± 0%   1.701µ ± 0%  -11.22% (p=0.000 n=10)
Unmarshal/CBOR_map_to_Go_map[int]any                         2.592µ ± 0%   2.553µ ± 0%   -1.49% (p=0.000 n=10)
Unmarshal/CBOR_map_to_Go_struct_keyasint                     1.825µ ± 0%   1.637µ ± 0%  -10.25% (p=0.000 n=10)
Unmarshal/CBOR_array_to_Go_[]any                             2.156µ ± 0%   2.097µ ± 0%   -2.76% (p=0.000 n=10)
Unmarshal/CBOR_array_to_Go_struct_toarray                    1.730µ ± 0%   1.558µ ± 0%   -9.89% (p=0.000 n=10)
UnmarshalCWTClaims                                           301.2n ± 0%   292.2n ± 0%   -3.02% (p=0.000 n=10)
UnmarshalSenML                                               1.280µ ± 0%   1.169µ ± 0%   -8.67% (p=0.000 n=10)
UnmarshalWebAuthn                                            376.2n ± 0%   341.2n ± 1%   -9.30% (p=0.000 n=10)
MapToStruct/default_options/all_known_fields                 385.4n ± 4%   328.6n ± 1%  -14.73% (p=0.000 n=10)
MapToStruct/default_options/all_known_dup_fields             297.8n ± 0%   258.2n ± 0%  -13.28% (p=0.000 n=10)
MapToStruct/default_options/all_unknown_fields               927.2n ± 6%   864.2n ± 5%   -6.79% (p=0.019 n=10)
MapToStruct/default_options/all_unknown_dup_fields           866.8n ± 2%   819.0n ± 2%   -5.52% (p=0.000 n=10)
MapToStruct/default_options/many_fields_one_key_per_field    2.804µ ± 3%   2.427µ ± 3%  -13.48% (p=0.000 n=10)
MapToStruct/reject_unknown/all_known_fields                  385.9n ± 4%   329.0n ± 2%  -14.75% (p=0.000 n=10)
MapToStruct/reject_unknown/all_known_dup_fields              297.2n ± 0%   258.2n ± 0%  -13.12% (p=0.000 n=10)
MapToStruct/reject_unknown/all_unknown_fields                207.7n ± 0%   207.3n ± 1%        ~ (p=0.528 n=10)
MapToStruct/reject_unknown/all_unknown_dup_fields            206.4n ± 1%   205.4n ± 1%        ~ (p=0.255 n=10)
MapToStruct/reject_dup/all_known_fields                      385.7n ± 4%   331.7n ± 2%  -14.00% (p=0.000 n=10)
MapToStruct/reject_dup/all_known_dup_fields                  210.8n ± 0%   207.5n ± 1%   -1.54% (p=0.000 n=10)
MapToStruct/reject_dup/all_unknown_fields                    1.536µ ± 3%   1.486µ ± 3%   -3.22% (p=0.009 n=10)
MapToStruct/reject_dup/all_unknown_dup_fields                386.6n ± 0%   379.6n ± 0%   -1.81% (p=0.000 n=10)
MapToStruct/reject_unknown_and_dup/all_known_fields          385.6n ± 4%   328.4n ± 1%  -14.82% (p=0.000 n=10)
MapToStruct/reject_unknown_and_dup/all_known_dup_fields      212.6n ± 0%   208.8n ± 1%   -1.76% (p=0.000 n=10)
MapToStruct/reject_unknown_and_dup/all_unknown_fields        206.9n ± 1%   206.4n ± 1%        ~ (p=0.148 n=10)
MapToStruct/reject_unknown_and_dup/all_unknown_dup_fields    205.8n ± 1%   205.3n ± 1%   -0.22% (p=0.041 n=10)

The next four benchmarks decode text strings that cannot take 
the new fast path for parsing, so they isolate its cost.

                                                            │     old     │                 new                 │
                                                            │   sec/op    │   sec/op     vs base                │
Unmarshal/CBOR_text_string_to_Go_interface_{}                104.1n ± 0%   105.6n ± 0%   +1.39% (p=0.000 n=10)
Unmarshal/CBOR_text_string_to_Go_string                      74.89n ± 1%   74.62n ± 1%        ~ (p=0.165 n=10)
Unmarshal/CBOR_indef-len_text_string_to_Go_interface_{}      509.7n ± 0%   505.7n ± 0%   -0.78% (p=0.000 n=10)
Unmarshal/CBOR_indef-len_text_string_to_Go_string            474.2n ± 0%   469.6n ± 0%   -0.95% (p=0.000 n=10)

To reduce scrolling, benchmark names were shortened:

• "indefinite-length" to "indef-len"
• "duplicate" to "dup"
• "UnmarshalMapToStruct" to "MapToStruct"

[x448]

🚀